[9fans] ipconfig DHCP problems

Akshat Kumar

unread,

Jul 9, 2010, 6:33:53 PM7/9/10

to

I've recently installed Plan 9 on a computer
with a 3COM 3C905 CX-TXM network card.
When I boot into the Plan 9 Live CD and run

ip/ipconfig

it communicates with the router and does
the whole DHCP business just fine.

However, when I boot into my newly installed
system and run

ip/ipconfig -D -d ether /net/ether0

I get:

ipconfig: dhcprecv: read timed out
...

forever. So, why would DHCP work in the
boot CD but not in the installed system?
The binaries haven't changed; I haven't
changed any hardware configurations.
What's going on?

Thanks,
ak

Akshat Kumar

unread,

Jul 9, 2010, 6:34:13 PM7/9/10

to

I will also add that specifying a static
address with

ipconfig -g gateway ether /net/ether0 ip mask

also does not work (trying ip/ping router,
for example, gives no communication).

So, still, what's going on?

ak

Akshat Kumar

unread,

Jul 9, 2010, 7:18:44 PM7/9/10

to

I added

*nomp=1
*nobiosload=1
*nodumpstack=1

to plan9.ini, and magically
DHCP worked.

Bugs? Hardware/BIOS
problems?

On Fri, Jul 9, 2010 at 3:28 PM, Akshat Kumar
<aku...@mail.nanosouffle.net> wrote:

erik quanstrom

unread,

Jul 9, 2010, 8:52:31 PM7/9/10

to

> *nomp=1
> *nobiosload=1
> *nodumpstack=1
>
> to plan9.ini, and magically
> DHCP worked.
>
> Bugs? Hardware/BIOS
> problems?

sounds like an interrupt routing problem. with
mp interrupts.

if you are using a 9atom kernel, the output of
/dev/irqalloc
/dev/mp*
/dev/kmesg

would be good to send along (offline) in my direction.
it sounds like the rx interrupt of your card is being
lost, and that would explain everything that you've
seen.

no (plan 9) bugs required.

- erik

erik quanstrom

unread,

Jul 9, 2010, 9:18:07 PM7/9/10

to

On Fri Jul 9 18:31:32 EDT 2010, aku...@mail.nanosouffle.net wrote:
> I will also add that specifying a static
> address with
>
> ipconfig -g gateway ether /net/ether0 ip mask
>
> also does not work (trying ip/ping router,
> for example, gives no communication).
>
> So, still, what's going on?

without knowing the specifics, it's very hard to say.
i suppose you did try ip/ipconfig without any options
at all? and that when you did try to specify everything
manually, you did use either an old-school netmask like
255.255.255.0 or /120 and not /24. since all plan 9 ip
addresses are stored in v6 format, the netmasks need
to specify the number of bits in the network accordingly.

when i debug dhcp problems, i generally start by running
snoopy on the dhcp server.

- erik

ron minnich

unread,

Jul 10, 2010, 12:07:28 AM7/10/10

to

When you start having interrupt problems, and they are fixed by nomp,
it's a good bet the motherboard has a bad MP table. Many, many boards
have tables that are quite broken. Vendors frequently change IRQ
hardware structure and the general rule is ACPI will likely be right,
because Windows wants it to be; PIRQ will likely be right; but for
some reason, _MP_ is many times quite wrong.

ron

erik quanstrom

unread,

Jul 10, 2010, 7:26:35 AM7/10/10

to

given an ioapic, the mp table just doesn't have enough information
to uniquely determine vector mappings. unfortunately this problem was
never solved. (vendor's answer: use acpi instead.) this is a leading
cause of "broken" mp tables. another leading cause is forgetting to
delete entries for hardware that bios has configured off.

- erik