[openib-general] Some Missing Features from mthca/user MAD access

[shaharf]

> Hello, Roland!
> Quoting r. Roland Dreier (roland.list at gmail.com) "Re: [openib-general]
> Some Missing Features from mthca/user MAD access":
> > > 1. The SM needs the ability to write (set/clear) the 'is_sm'
PortInfo
> > > capability mask bit (from user space).
> >
> > Yes, I'll implement this.
> 
> Didnt Shahar already do this?

No I didn't. The gen2 opensm doesn't set the IS SM bit for now.

> Anyway, as a reminder, there are a couple of issues to consider with
this:
> 1. there shall, properly, be only one sm at the same time
> 
>    so I think an attempt to set it when its already set
>    shall fail or block.
> 
> 2. when sm exits the bit shall be auto-cleared
> 
>    I guess the proper way to do it is on closing the file.
> 
> HTH
> mst

I agree with Michael that it would be nice of the kernel to coordinate
this and to clean it automatically if (and when) the SM fails. 

How ever, to Roland's question, no, it is not critical. I can try to
clean it up when exiting. If the process will be killed, the cleanup
process will not happen. This is not critical, but it may delay other SM
from taking over. They would attempt to ping the node with SMINFO until
they timeout (SM depended) and then they would take over. This may
repeat every heavy sweep, so it would be nice to have some sort of
cleaning mechanism.

I am not sure what is so bad with having a file controlling it (and if
it is closed, cleaning it automatically).

Alternate cleaning mechanisms may be a small local agent that will probe
(using sminfo) the local SM and will clean this bit once it stops
responding. This is feasible (a small script will do), but I am not sure
it is cleaner then a sysfs file.

Shahar