Re: [Freeipmi-users] Serial Console hang during reboot

[Al Chu]

Hey Phil,

On Sat, 2011-08-27 at 12:55 -0700, Phil Pokorny wrote:
> I wonder if the BMC internal Linux kernel is generating a HUP on
> carrier detect changing during the reboot and killing the SOL process
> inside the BMC.
>
> The plan was to use a console monitoring/aggregation tool like Conman
> or Conserver so we'll look into implementing serial keepalive there.

I'm not 100% sure, but I think the newest conman allows you to pass
workaround flags in via the conf file.  Can't speak to conserver.

> Is there a non-data packet you could send (like telnet are-you-there)
> or a duplicate tcp ack that you could send instead of nul bytes that
> make their way into the input stream?

To my knowledge no.  The reason I implement it as serial NUL chars, is
because the IPMI session appears to be completely alive and fine.  e.g.
when libipmiconsole sends IPMI "pings" to keep the session alive, the
BMC ACKs them.  So I didn't think anything within the IPMI protocol
would work.  It had to be something within the serial data stream to
detect the problem.

However, this option has not be ridiculously tested and I didn't have a
motherboard to play with on this.  It was my initial first guess on how
to solve the problem.  There could totally be something more reasonable.
Thinking about it a minute, I suppose it's possible I could try sending
a serial break or some similar data outside of the data stream.

LMK if serial-keepalive works for you.  If it doesn't, we can try and
play with some other tricks.

Al


> Thanks,
> Phil P.
> 
-- 
Albert Chu
address@hidden
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory