[OpenIndiana-discuss] Serial console for OI/Solaris on Intel MFSYS blades

[Jim Klimov]

2012-06-14 18:53, Jim Klimov пишет:
> Hello all,
>
>    I have a peculiar problem: the Intel MFSYS blade servers, in short ;)
>
>    They have a number of interesting features, due to which are deployed
> at a number of our customers, but with somewhat under-cooked firmware
> features wherever I've seen. The one bugging me this week is a Remote
> Serial Console implementation.
>
>    This blade chassis has up to 6 server blades, a management module,
> 1 or 2 storage controller modules driving the 14 disks (the SCMs do
> the RAID and publish LUNs to server blades), and 1 or 2 networking
> modules (GbE switches). Overall, quite a bit of redundancy would
> make it a good box, if its code was all working properly ;)
>
>    So, the management module takes the servers' redirected text console
> IO and publishes that as a "SOLCLI" (Serial-Over-LAN CLI) session via
> SSH on dedicated ports like 2201-2206 for the 6 blades. I found next
> to zero specific info on the serial emulation, but it seems to be
> either auto-bauding or (as the Linux blades successfully use it) a
> "115200,8,n,1,-" link. I'd bet auto-bauding, because "9600,8,n,1,-"
> also seems to work. Terminal emulation is unknown, but somewhat like
> VT100 (however, GRUB menu fails to render properly, unless I specify
> "--dumb" for a simplified rendering mode).
>
>    The problem is, the redirected console gets lost after Solaris 10
> or OI boots on the blade. It works somewhat for BIOS and GRUB (but
> even there it sometimes loses connections or farts out invalid
> characters), and if I boot with "-B console=ttya" - sercon usually
> makes it through the OS build version banner and prints a character
> or two of the OS initialization (console logs from SMF startups).

Just in case, here's all the text that makes it to the console:

========
OpenIndiana Build oi_151a4 64-bit (illumos 13676:98ca40df9171)
SunOS Release 5.11 - Copyright 1983-2010 Oracle and/or its affiliates.
All rights reserved. Use is subject to license terms.
WARNING: No randomness provider enabled for /dev/random. Use 
cryptoadm(1M) to enable a provider.
Mo
========

("Mo" seems to be part of the messages reporting the mounting
of root dataset components, I added debugging to my SMF scripts
and that should not be so fatal).

Thanks to Gary's pointers, I looked at "asy" driver code and
found a few tunables (to disable or enforce some tests, and
to limit the FIFO size). I've set them into /etc/system, and
they do get applied upon boot, but they did not help too :(

*** 
http://src.illumos.org/source/xref/illumos-gate/usr/src/uts/common/io/asy.c
set asy:asy_max_tx_fifo=1
set asy:asy_fifo_test=0
set asy:asy_scr_test=0

However, while playing around when the system was booted with
the serial console as the system one, restarting the console
ttymon (svcadm restart console-login:default) has led to the
cached characters (of that SMF script debug output) being
output from the serial console, one character per ttymon stop,
unless I restart them too quickly. Now, that is kind of weird,
but it is something for us to follow on... (and for volunteers
to elaborate on) ;)

Thanks,
//Jim