Message-ID: <20090819185605.GA19345@kaos.fbx.proxad.net>
Date:	Wed, 19 Aug 2009 20:56:05 +0200
From:	Benoît Vaillant <benoit.vaillant@...log.org>
To:	netdev@...r.kernel.org
Subject: sd8686 linux system hang: not always

Hi list,

It seems my previous message for libertas-dev didn't make its ways
through, so resending it... Appologies if it did in any way and for
multiple receptions if any :/

Here are a few notes, to try to isolate the issue(s?) regarding this
problem.

/*
Just a bit of context information:
I'm using an MID (quite close to an Aigo, although bios differs,
probably some hardware too, yet I've not checked much on that). I'm
trying to use my Marvell card to get wifi access on a debian
installation. The other OS installed (midinux) successfully gets a
connection using the 8686_v9 mobilin driver, so I'd not go for any
hardware issue. This driver failing to compile 'out of the box' on
fresh kernels, I thought getting the libertas module working would be
a better way to get through.
*/

I had the same kernel hang issue as reported earlier on this list[1]
and can also confirm the reload issue[2] still seems to be here,
although I'm not really bothered by that for the moment and didn't
take any time at assessing it.

I did a few attempts at getting associated with an AP, initially
unsuccessfully getting a hang, my latest attempts leading to a better
result (no association, yet no hang! :) ).

I've tried both 8.73.7.p3 and 9.70.3.p24 versions of the firmware (and
the one provided by midinux, which I highly suspect to be a mere
8.73.7.p3). In my latest attempts, I focused only on the 8.73.7.p3
since it gave good results using iwlist scannig: I could list the
available APs fine.

Now, with the attempts I did, using a kernel 2.6.31-rc5, I used to
hang. Using some manually inserted printk()s, it lead me from the
libertas driver to mmc, esp. in drivers/mmc/core.c the
mmc_wait_for_req(...) function. It was done without mmc debugging
messages, with libertas debugging symbols activated but not printed
out.

>From there on, I switched to the wireless git[3] in order to be able
to provide more convinient patches if needed. I immediately patched it
against the attached proposal from last thread[4], adapting it due to
compile issues (see below). I also gave a quick try at Philip's
proposal[5], which didn't seem to work, unfortunately.

about the patch: I didn't adapt it as proposed latter on in the thread
by setting 0 values, but by reinserting the "old"
cmd_ds_802_11_get_log log structure. Wrong thing?

The good news using this kernel was that I wasn't always stuck in a
hang: listing the APs works, connecting to an AP using WPA & french
bands failed, yet without hanging (here, I had MMC & libertas
debugging messages).

The bad news is I can reproduce the hang using these sources quite
easily if I do, say: an association request, followed by an iwconfig
to check if the association was done *before* all the attempts were
done. The attempts I did using iwconfig once the previous one
timed-out were successfull.

This leads me to think there are in fact two issues:
 - an association issue (at least in France). Maybe not much of a
   problem? I might even be the one doing the wrong thing, not the
   kernel :)
 - an issue when accessing the card (race condition?), which could be
   more subtle to handle.

Using the fresh 2.6.31-rc6 kernel.org release, with MMC & libertas
debugging messages on, I still can do an association request without
getting neither a hang, nor an association. So maybe one of the race
issues that are patched in it do solve part of this problem. Yet,
running several AP association requests does hang.

I'd be glad to do more testing and provide usefull information. Any
idea(s) on which track(s) I should now follow and how I can help you
solve this issue?

Thanks,

-- 
Benoît

[1] http://lists.infradead.org/pipermail/libertas-dev/2009-June/002487.html
[2] http://lists.infradead.org/pipermail/libertas-dev/2009-June/002510.html
[3] git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-testing.git
[4] http://lists.infradead.org/pipermail/libertas-dev/2009-June/002496.html
[5] http://lists.infradead.org/pipermail/libertas-dev/2009-June/002494.html
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists