too many packets lost (t40p with wlan)

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

Hi Folks,

finally I mananged connecting my notebook (t40p) with 2.6.3 via madwifi-driver to my router (incl. WEP).
but when i ping my router I get the following output:

ben · Posted: Wed Mar 10, 2004 8:18 pm Post subject:

Hi,

If you use ACPI and you loaded processor.o you could have this problem. You could try to unload processor and then reload it.

what gives dmesg?

Do you use the cvs version of madwifi?

Ben

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

Hi Ben,

I have already read something about troubles from acpi concerning wlan. and indeed i have
acpi turned on [it's a laptop

]. Nevertheless I'll give it a try.
output from dmesg

ben · Posted: Thu Mar 11, 2004 7:16 am Post subject:

Hi Pinky,

First, you could have a look at my page http://www.ontheedge.ch/acpi.html
which describes my probe with acpi. Basically, if I recall correctly, the processor module (you did make it a module, right?) interfere badly with madwifi. You should see a lot of ath0 resetting: hardware errors in dmesg.

I use madwifi-cvs because of the RFMON capabilities (wardriving) that where not present in the ebuild in december-january.

I really think that the processor.o module is causing the problem, but it could also be some networking problem, so how does your routes look likes (netstat -nr) ?

HTH

Ben

P.S. I know the t40p is a laptop, I have the same wonderfull machine
http://www.ontheedge.ch/t40p.html

Arg, edit:

the processor module is almost only used to get the temperature, so even if you do want to stick with acpi (In my opinion apm is much better for the moment, for this laptop, suspend to ram works, battery life is much longer...), you can have all speedsted you want without processor.o.
Then if you rmmod processor.o, the kernel will oops, and then you can modprobe it again, and everything works alright, e.g. madwifi and temperature.

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

Hi Ben,

I had acpi compiled into my kernel. but this should not be the problem. yesterday I built one without any acpi support and I have the exact same problem.

sorry, what is wardriving?

and here my netstat -nr output

ben · Posted: Thu Mar 11, 2004 8:27 pm Post subject:

Ah ha, so, you are the *one* who read my page !

Wardriving is more or less the ability to sniff wireless data. Basically you need to put the interface in monitoring mode to be able to see the raw packet. (Actually it means driving around to find hot spot you can connect to)

Back to your problem: do you know if you get the same problem with a wired link (standart ethernet).? I had some problem with another computer of mine because of a race condition between two interrupts. Since then, I disable APIC (yes this time it is APIC, not ACPI) in the BIOS, and I make sure that I don't have smp support on uniprocessor.

HTH

Ben

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

oh, forgot to mention:

the problem only occurs with wlan
eth runs smoothly.

pinky

ben · Posted: Thu Mar 11, 2004 8:53 pm Post subject:

Sorry that I can't help you much more. Some idea to go further:

Try without WEP to see what happened

diff your kernel config and mine to see the difference (everything related to ACPI, SMP, net)

Good luck

Ben

BTW, it really look like there is a conflict between processor.o and madwifi. Are you sure to run the kernel without ACPI? (I know, I know, but I have to ask)

rcast · n00b Joined: 22 Apr 2003 Posts: 39

Just a note,
the hardware reset error that you get with the madwifi driver only occur when the processor is in the c3 state. Running xmms in the background does seem to stop this and stop the hardware reset error.

amoc · Posted: Tue Apr 20, 2004 5:31 pm Post subject:

I think it has something to do with WEP. When I use my T40p without WEP it works just fine. Haven't found a solution tho.

I know about the issue concerning cvs and ebuilds, but I think there should be a cvs ebuild for madwifi in this case.

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

@ rcast

I've tried cvs-madwifi as well (without success). the newest version, I checked out today, even worsens performance. Sorry, but I couldn't catch your idea about xmms. Could you explain it?

amoc · Posted: Tue Apr 20, 2004 6:18 pm Post subject:

Yes, I believe it's the WEP. I have another laptop in the same WLAN that works great with another driver (different card).

When I was visiting a friend recently I accessed his WLAN without WEP and it was rock stable (I was connected for several days) using the 0.1_pre20040212 ebuild.

rcast · n00b Joined: 22 Apr 2003 Posts: 39

I have npt yet experienced any problems using WEP and madwifi, though i only used it a few days while i was at home. At unni they don't have wep enabled so i can't say much about that.

The tip that i gave about xmms is something i read on a fourum which preveneted the processor from going into the C3 powersaving mode.

if

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

amoc · Posted: Wed Apr 21, 2004 8:33 am Post subject:

I'm not using ACPI (I use APM) and I'm still having the same problem with my T40p.

rcast · n00b Joined: 22 Apr 2003 Posts: 39

Sorry, can't help you much then.
Though you might want to ocnsider using driverloader(worked without a single issue) or ndiswrapper (personally i never got this to work)

Rene

Pinky · n00b Joined: 05 Nov 2003 Posts: 18

Hi folks, I could finally solve (at least) my problem.

while windows was resistant to some router (mis)settings, linux seems to be more sensitive. I had to reduce both, my router's BASIC and TX rate to 11Mbs. Somehow the TX rate were set to 22Mbs.

but nevertheless thx a lot for your patience and ideas.

Pinky

ben · Posted: Wed Apr 28, 2004 1:45 pm Post subject:

I just wanted to add my voice too:

I *HAD* madwifi from cvs working very well with kernel 2.6.1 with apm (even with acpi, but without processor module loaded).

Then I upgraded to kernel 2.6.5 and now I have the same problem with acpi and with apm: ath0 is choppy under load, althoug it seems to work a bit better with acpi. Too bad acpi is not good at power consumption during a sleep (around 10 hours instead of many days under apm)

So every input would be appreciated.

Ben

ben · Posted: Mon May 03, 2004 7:57 am Post subject:

Well, some news about this:

As I said, it used to work with the cvs version of around january under apm and acpi provided processor was not loaded (with acpi).

Then I upgraded to kernel 2.6.5 and also to the latest cvs. Madwifi would then throw ath0: transmit timed out and resetting hardware every minutes or so, especially under load. This happened with apm and with acpi with and without processor loaded.

So I learned that:
1.- REGPARM should not be set in the kernel (conflict with proprietary modules)
2.- the madwificvs from early 2004 works ok under apm, and may be dependant of C2 under acpi
3.- the latest madwifi-cvs does reset (timed out) under both apm and acpi. This one has the nifty feature of lightening the wifi led. It was compiled with and without COPTS= -NOWME. Still the 4 queues appeared in dmesg.

Conclusion:
For me, my t40p work very well with apm, kernel 2.6.5, REGPARM not set, madwifi-cvs from early 2004 (but not the latest).

HTH

Ben

P.S. I really can't wait for the time acpi will be fully supported (meaning that the laptop can really suspend to ram (power consumption), and can properly resume (especially the usb subsystem)). In the mean time I'll stay with apm.

rab · n00b Joined: 23 Dec 2003 Posts: 19 Location: Switzerland

Hello ben

How can I get madwifi-cvs "but not the latest"?
(i don't know cvs at all

)

rab

ben · Posted: Thu May 13, 2004 3:51 pm Post subject:

hi,

Bad news I don't know either. I had a copy from that time :-)

It is something like cvs -D date. Maybe man cvs could help.

Good news. The very latest cvs is or will real soon now corrected. Atheros_greg released a hal.o which is ok.
You may be able to find it in the archive of madwifi mailind list in case it is not already in cvs.

HTH

Ben