[openib-general] RE: [PATCH] osm: support for trivial PKey manager

[Ofer Gigi]

Hi Hal,
Sorry, I didn't see there are other questions below....

Here are the answers:
1. Might 0x7fff be better in this case ?
As I said, currently all IB devices come up from reset with
preconfigured 0xffff key. Since we don't want to be inconsistent, we
decided to insert 0xffff.
2. Can you elaborate on any plans in this area as to timing and what
"full"
means ?
Since all IB devices come up from reset with 0xffff, and anyway, we
insert it when 0x7fff or 0xffff doesn't exist, currently all physical
ports belong to the same Pkey group, so there is no meaning to pkey
groups.
In the future, we would like to have a Pkey manager that will manage
different Pkey groups (that don't share the same pkeys) for the physical
ports.
3. Can you elaborate on the testing ?
In the test, the following steps were done:
1. Remove 0x7fff or 0xffff from all physical ports
2. Force full sweep.
3. Verified that all physical ports have now 0xffff

Thanks!
Ofer     


-----Original Message-----
From: Hal Rosenstock [mailto:halr at voltaire.com] 
Sent: Thursday, December 29, 2005 4:24 PM
To: Ofer Gigi
Cc: OPENIB
Subject: Re: [PATCH] osm: support for trivial PKey manager

Hi Ofer,

On Thu, 2005-12-29 at 05:20, Ofer Gigi wrote:
> Hi Hal,
> 
> My name is Ofer Gigi, and I am a new software engineer in Mellanox 
> working on OpenSM.  
> This patch provides a new manager that solves the following problem:

Thanks! I am starting to work on integrating this. 

I have a couple of questions and some comments and questions embedded
below.

How are the PKeys configured ?

Does the SM always go through the extra states to set PKeys ? What if
there are none to be set ?

> OpenSM is not currently compliant to the spec statement:
> C14.62.1.1 Table 183 p870 l34:
> "However, the SM shall ensure that one of the P_KeyTable entries in
every
>  node contains either the value 0xFFFF (the default P_Key, full
membership) 
>  or the value 0x7FFF (the default P_Key, partial membership)."
> 
> Luckily, all IB devices comes up from reset with preconfigured 0xffff
key.
> This was discovered during last plugfest. 

> To overcome this limitation I implemented a simple elementary PKey
manager
> that will enforce the above rule (currently adds 0xffff if missing).

Might 0x7fff be better in this case ?

> This additional manager would be used for a full PKey policy manager 
> in the future.

Can you elaborate on any plans in this area as to timing and what "full"
means ?

> We have tested this patch 

Can you elaborate on the testing ?

Thanks again.

-- Hal