2.6.24-rc8: iwl3945 gets stuck

2.6.24-rc8: iwl3945 gets stuck

[Harald Dunkel]

Hi folks,

If I put some heavy load on the iwl3945, then the network connection
gets stuck after a some time. To fix it I have to reload the module.

AFAICS this problem was a topic on lkml almost 3 months ago. Any news
about this? I would be glad to help to track this down, but I have
no idea how to change the scaling algorithm to iwl-3945-rs .


Regards

Harri

Re: 2.6.24-rc8: iwl3945 gets stuck

[John W. Linville]

On Tue, Jan 22, 2008 at 09:54:11PM +0100, Harald Dunkel wrote:

> If I put some heavy load on the iwl3945, then the network connection
> gets stuck after a some time. To fix it I have to reload the module.

Can you quantify this a bit more?  What constitutes a "heavey load"?
What (if any) encryption are you using?  Are you using any options
for iwl3945 in /etc/modprobe.conf?

Could you include the output of dmesg and/or the contents of
/var/log/messages (trimmed for the most recent boot)?

> AFAICS this problem was a topic on lkml almost 3 months ago. Any news
> about this? I would be glad to help to track this down, but I have
> no idea how to change the scaling algorithm to iwl-3945-rs .

This should happen automatically now.

John
-- 
John W. Linville
linville@tuxdriver.com

Re: 2.6.24-rc8: iwl3945 gets stuck

[Daniel Hazelton]

On Tuesday 22 January 2008 17:15:42 John W. Linville wrote:
> On Tue, Jan 22, 2008 at 09:54:11PM +0100, Harald Dunkel wrote:
> > If I put some heavy load on the iwl3945, then the network connection
> > gets stuck after a some time. To fix it I have to reload the module.
>
> Can you quantify this a bit more?  What constitutes a "heavey load"?
> What (if any) encryption are you using?  Are you using any options
> for iwl3945 in /etc/modprobe.conf?
>
> Could you include the output of dmesg and/or the contents of
> /var/log/messages (trimmed for the most recent boot)?
>
> > AFAICS this problem was a topic on lkml almost 3 months ago. Any news
> > about this? I would be glad to help to track this down, but I have
> > no idea how to change the scaling algorithm to iwl-3945-rs .
>
> This should happen automatically now.
>
> John

I've been getting a warning in the dmesg of my laptop with every boot since I 
started using 2.6.24-rc7 that might be related.

This doesn't appear to cause any problems, but from looking at the source 
of the warning it appears that the ipw3945 hardware might be causing the 
problem.
 
[   31.460143] ADDRCONF(NETDEV_CHANGE): eth1: link becomes ready
[   31.549722] WARNING: at net/mac80211/rx.c:1486 __ieee80211_rx()
[   31.549817] Pid: 4436, comm: amixer Not tainted 2.6.24-rc7-git2 #1
[   31.549903]
[   31.549904] Call Trace:
[   31.550063]  <IRQ>  [<ffffffff8823c309>] :mac80211:__ieee80211_rx+0xc99/0xd60
[   31.550236]  [<ffffffff80473a36>] _spin_unlock_irqrestore+0x16/0x40
[   31.550332]  [<ffffffff8828930a>] :iwl3945:iwl_rx_queue_restock+0xca/0x170
[   31.550422]  [<ffffffff80473a36>] _spin_unlock_irqrestore+0x16/0x40
[   31.550520]  [<ffffffff8822d228>] :mac80211:ieee80211_tasklet_handler+0xb8/0x120
[   31.550646]  [<ffffffff80246741>] tasklet_action+0x51/0xc0
[   31.550732]  [<ffffffff80473974>] _spin_unlock+0x14/0x40
[   31.550820]  [<ffffffff80246644>] __do_softirq+0x64/0xe0
[   31.550909]  [<ffffffff8020d57c>] call_softirq+0x1c/0x30
[   31.550995]  [<ffffffff8020ef0d>] do_softirq+0x3d/0x90
[   31.551083]  [<ffffffff80246558>] irq_exit+0x88/0xa0
[   31.551169]  [<ffffffff8020f025>] do_IRQ+0xc5/0x1b0
[   31.551257]  [<ffffffff8020c8d1>] ret_from_intr+0x0/0xa
[   31.551369]  <EOI>  [<ffffffff802874be>] get_page_from_freelist+0x30e/0x670
[   31.551519]  [<ffffffff802878ce>] __alloc_pages+0x6e/0x3b0
[   31.551608]  [<ffffffff80283417>] generic_file_aio_read+0xd7/0x180
[   31.551699]  [<ffffffff802a256c>] alloc_page_vma+0x9c/0xf0
[   31.551788]  [<ffffffff8029281e>] handle_mm_fault+0x50e/0x780
[   31.551874]  [<ffffffff80473974>] _spin_unlock+0x14/0x40
[   31.551962]  [<ffffffff80473a36>] _spin_unlock_irqrestore+0x16/0x40
[   31.552052]  [<ffffffff80475a68>] do_page_fault+0x228/0x970
[   31.552146]  [<ffffffff80473974>] _spin_unlock+0x14/0x40
[   31.552251]  [<ffffffff802b2b5e>] vfs_read+0x13e/0x180
[   31.552340]  [<ffffffff80473cc9>] error_exit+0x0/0x51
[   31.552436]

The location of the warning is:
        hdrlen = ieee80211_get_hdrlen(rx.fc);
line in question -->        WARN_ON_ONCE(((unsigned long)(skb->data + hdrlen)) & 3);

        if (type == IEEE80211_FTYPE_DATA || type == IEEE80211_FTYPE_MGMT)
                local->dot11ReceivedFragmentCount++;

        sta = rx.sta = sta_info_get(local, hdr->addr2);

Now, the problem is that this might be nothing, and it might be the cause 
of the problem. (I don't think it is the cause, myself, because I've subjected
my laptop to a lot of activity - to the point that the card was starting to drop
packets - and have seen no problems)

DRH

--
Dialup is like pissing through a pipette. Slow and excruciatingly painful.

Re: 2.6.24-rc8: iwl3945 gets stuck

[Harald Dunkel]

Jeff suggested to post this on linux-wireless@vger.kernel.org.
I hope this is OK?


Regards

Harri

2.6.24-rc8: iwl3945 gets stuck

[Harald Dunkel]

Hi folks,

I've got some problem with iwl3945: If I run

	xset dpms force off

for the local display while some network traffic is running, then
the iwl3945 gets stuck. Syslog says:

Jan 23 19:29:41 daffy kernel: iwl3945: Microcode SW error detected.  Restarting 0x2000008.
Jan 23 19:29:43 daffy kernel: iwl3945: Can't stop Rx DMA.

The same happens if the Xserver runs into the configured timeout.
If the network is idle, then there is no such problem, as it seems.
If the Xserver is not running, and if the console runs into dpms,
then iwl3945 seems to stay alive, too, but to be sure I would need
some command similar to the xset shown above.

If the net is stuck (and if I am lucky), then I can rmmod iwl3945 and
load it again, but sometimes I have to reboot.

The machine is a Dell laptop xps m1330 running 64bit kernel 2.6.24-rc8.
Encryption is wpa/psk. iwl3945 is loaded without any special options.
At boot time there were no unusual call traces or something like this.

The configured XWindow driver is "nv", but I could reproduce this with
NVidia's proprietary "nvidia" driver, too. The vga device is a "nVidia
Corporation GeForce 8400M GS".

I would be glad to help to track this down.


Regards

Harri

RE: 2.6.24-rc8: iwl3945 gets stuck

[Chatre, Reinette]

On ,  Harald Dunkel wrote:

> I've got some problem with iwl3945: If I run
> 
> 	xset dpms force off
> 
> for the local display while some network traffic is running, then
> the iwl3945 gets stuck. Syslog says:
> 
> Jan 23 19:29:41 daffy kernel: iwl3945: Microcode SW error
> detected.  Restarting 0x2000008.
> Jan 23 19:29:43 daffy kernel: iwl3945: Can't stop Rx DMA.
> 

Very strange. Would it be possible to load the module with
"debug=0x43fff"? This will capture a detailed event log and error log
from uCode, so we can see what caused the uCode error.

Also, could you please open a bug in our bug tracker
(http://www.bughost.org/bugzilla/) - this will make it easier to keep
all info together and get the right people to take a look at it.

Thanks

Reinette

Re: 2.6.24-rc8: iwl3945 gets stuck

[Harald Dunkel]

Chatre, Reinette wrote:
> 
> Very strange. Would it be possible to load the module with
> "debug=0x43fff"? This will capture a detailed event log and error log
> from uCode, so we can see what caused the uCode error.
> 

I will try. Is there some way to turn debug on and off after
the module has been loaded, e.g. by running

	echo 0x43fff >/proc/some/fancy/path

?


Regards

Harri

Re: 2.6.24-rc8: iwl3945 gets stuck

[Joonwoo Park]

On Sat, Jan 26, 2008 at 03:54:30PM +0100, Harald Dunkel wrote:
>
> I will try. Is there some way to turn debug on and off after
> the module has been loaded, e.g. by running
>
> 	echo 0x43fff >/proc/some/fancy/path
>
> ?

please build kernel with CONFIG_IWLWIFI_DEBUG=y
mount sysfs on /sys
echo 0x43fff > /sys/bus/pci/drivers/iwl3945/debug_level

Thanks,
Joonwoo

Re: 2.6.24-rc8: iwl3945 gets stuck

[Harald Dunkel]

Chatre, Reinette wrote:
> 
> Very strange. Would it be possible to load the module with
> "debug=0x43fff"? This will capture a detailed event log and error log
> from uCode, so we can see what caused the uCode error.
> 
> Also, could you please open a bug in our bug tracker
> (http://www.bughost.org/bugzilla/) - this will make it easier to keep
> all info together and get the right people to take a look at it.
> 

moving to bug tracker, see http://bughost.org/bugzilla/show_bug.cgi?id=1577 .
The requested debug output has been posted there, too.


Regards

Harri