[ltp] ibmtr_cs 2.2.18 and DHCP

[Burt Silverman]

Markus Alt writes:

>Usually, I use a fixed IP address for my machine. Last week I had to
>connect my TP to another location's LAN and therefore use DHCP for the
>IP configuration. To my surprise, this wasn't working, although I had
>used DHCP without problems before. Today, I double-checked it in our LAN
>with the same unsatisfying result.

>It looks like the adapter isn't opened correctly and the tr0 interface
>receives a lot of packages, but is not able to send some.

Markus,

I have to believe once again that this is a Ring Parameter Server problem;
that bad RPS always sets the Frame Control field FC to 0 rather than to
0x01 in all of its responses to the adapter, and the adapter cannot open
reliably. Imagine my frustration sitting at IBM's main site in Research
Triangle Park and waiting on baited breath (bated breath is perhaps the
usually correct expression) for network maintainers to acknowledge the
problem. Note that I said acknowledge. Experience tells me that the
acknowledgement phase is the longest; if they acknowledge, they may repair
reasonably quickly. Hey, imagine also if we never moved and I never had a
bad RPS, I'd have zero suggestions!

The opening of the adapter onto the ring is independent of the driver --
once we kick off the command, the adapter does everything on its own.
Perhaps one caveat would be that packets are not being sent to the driver
for transmission. For my own piece of mind, I just verified that Linux
enforces this. It does so by not calling dev_activate() [in
net/sched/sch_generic.c] until the open has successfully completed.

So rest assured, your adapter not opening is most likely because of a
misconfigured RPS.

Having said that, I also suspect that my driver is less forgiving of the
bad RPS than previous drivers, because I merely close and reopen the
adapter; the old driver did a complete reinitialization. I don't know for
certain, but the complete reinit may do a better job of guaranteeing that
all adapter buffers are clean. So that even if it takes more than one try
to open on the ring, you have a better chance the second time of success.
This is just speculation.

I have delivered another driver to the Linux Token Ring Project. After the
holiday, I may compare it with the current driver on the bad RPS ring.

>I have *not* disabled (i.e.
>commented out) the options for the ibmtr_cs module within
>/etc/pcmcia/config.opts, although some time ago Burt told us to do so
>and let the driver pick the values. The reason is that my machine hung
>from time to time when not specifying these options for ibmtr_cs.

Perhaps if you trap the problem, then send me the /var/log/messages and
/etc/pcmcia/config.opts, I can see if there is something obvious.

Burt

ps Get ibmtr_cs.c from pcmcia-cs-3.1.23 for extra robustness, but it won't
solve this problem.

----- The Linux ThinkPad mailing list -----
The linux-thinkpad mailing list home page is at:
http://www.bm-soft.com/~bm/tp_mailing.html