rmmod deadlocks with 2.6.5-rc[2,3]

rmmod deadlocks with 2.6.5-rc[2,3]

[Frédéric L. W. Meunier]

If I boot with 2.6.5-rc[2,3] and use rmmod snd_via82xx or rmmod
ohci_hcd (it doesn't happen with all modules), rmmod deadlocks.
2.6.4 works fine.

-- 
http://www.pervalidus.net/contact.html

Re: rmmod deadlocks with 2.6.5-rc[2,3]

[Daniel Drake]

Hi,

Frédéric L. W. Meunier wrote:
> If I boot with 2.6.5-rc[2,3] and use rmmod snd_via82xx or rmmod
> ohci_hcd (it doesn't happen with all modules), rmmod deadlocks.
> 2.6.4 works fine.

The ohci_hcd problem should be temporarily fixed by a recent patch to this 
list, from Greg KH (subject: [PATCH] USB: Eliminate wait following interface 
unregistration). This worked for me.

As for the snd_ modules, this is a different problem, which I am still 
experiencing. I have had it with both snd_emu10k1 and snd_intel8x0 - but it 
does not happen every time. I have experienced it on 2.6.5-rc2 and -rc3 (plus 
their -mm patches). rmmod hangs and doesnt respond to kill -9.

Is there any output I can capture to diagnose this?

Daniel

Re: rmmod deadlocks with 2.6.5-rc[2,3]

[Frédéric L. W. Meunier]

On Tue, 30 Mar 2004, Daniel Drake wrote:

> Frédéric L. W. Meunier wrote:
> > If I boot with 2.6.5-rc[2,3] and use rmmod snd_via82xx or rmmod
> > ohci_hcd (it doesn't happen with all modules), rmmod deadlocks.
> > 2.6.4 works fine.
>
> The ohci_hcd problem should be temporarily fixed by a recent
> patch to this list, from Greg KH (subject: [PATCH] USB:
> Eliminate wait following interface unregistration). This
> worked for me.

I forgot to report that I don't need OHCI. It's hotplug which
loads it for some reason under 2.6, but not under 2.4.

> As for the snd_ modules, this is a different problem, which I
> am still experiencing. I have had it with both snd_emu10k1
> and snd_intel8x0 - but it does not happen every time. I have
> experienced it on 2.6.5-rc2 and -rc3 (plus their -mm
> patches). rmmod hangs and doesnt respond to kill -9.

I used ALSA from CVS.

> Is there any output I can capture to diagnose this?

Here nothing gets printed. Maybe strace can help. But I'll
wait.

It also happened removing i2c_isa when I went through removing
all modules. Nothing wrong with joydev, adi, gameport, it87...

-- 
http://www.pervalidus.net/contact.html

udev related ? (Was Re: rmmod deadlocks with 2.6.5-rc[2,3])

[Frédéric L. W. Meunier]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 2196 bytes --]

OK, I gave 2.6.3-rc3-mm1 a try and could reproduce it.

I still compile with CONFIG_DEVFS_FS=y and
CONFIG_DEVFS_MOUNT=y, but to use udev boot with devfs=nomount.

I then booted with devfs=mount acpi=off noapic. Guess what ? I
was able to unmount all offending modules.

What next ? Boot with devfs=mount to see if ACPI was to blame.
No, it wasn't.

My /etc/rc.d/rc.S is very simple:

[ -e /dev/.devfsd -a -x /sbin/devfsd ] && devfsd /dev

(only run devfsd if devfs is mounted)

mount -vn -t proc proc /proc # Needed for LABEL= in /etc/fstab

mount -vn -t sysfs sysfs /sys

[ ! -e /dev/.devfsd -a -d /sys/block ] && /etc/rc.d/start_udev

(only run start_udev if devfs isn't mounted and sysfs is
mounted)

Since 2.6.4 works fine for me using devfs=nomount, something is
broken. I'm using the latest udev and hotplug.

I'm attaching a strace of a rmmod that failed.

On Tue, 30 Mar 2004, Frédéric L. W. Meunier wrote:

> On Tue, 30 Mar 2004, Daniel Drake wrote:
>
> > Frédéric L. W. Meunier wrote:
> > > If I boot with 2.6.5-rc[2,3] and use rmmod snd_via82xx or rmmod
> > > ohci_hcd (it doesn't happen with all modules), rmmod deadlocks.
> > > 2.6.4 works fine.
> >
> > The ohci_hcd problem should be temporarily fixed by a recent
> > patch to this list, from Greg KH (subject: [PATCH] USB:
> > Eliminate wait following interface unregistration). This
> > worked for me.
>
> I forgot to report that I don't need OHCI. It's hotplug which
> loads it for some reason under 2.6, but not under 2.4.
>
> > As for the snd_ modules, this is a different problem, which I
> > am still experiencing. I have had it with both snd_emu10k1
> > and snd_intel8x0 - but it does not happen every time. I have
> > experienced it on 2.6.5-rc2 and -rc3 (plus their -mm
> > patches). rmmod hangs and doesnt respond to kill -9.
>
> I used ALSA from CVS.
>
> > Is there any output I can capture to diagnose this?
>
> Here nothing gets printed. Maybe strace can help. But I'll
> wait.
>
> It also happened removing i2c_isa when I went through removing
> all modules. Nothing wrong with joydev, adi, gameport, it87...

-- 
http://www.pervalidus.net/contact.html

[-- Attachment #2: Type: APPLICATION/octet-stream, Size: 950 bytes --]

Re: udev related ? (Was Re: rmmod deadlocks with 2.6.5-rc[2,3])

[Greg KH]

On Tue, Mar 30, 2004 at 07:49:11PM -0300, Fr?d?ric L. W. Meunier wrote:
> OK, I gave 2.6.3-rc3-mm1 a try and could reproduce it.

Yes, the bug is still there.  It is being activly worked on as I type
this...

And no, it's not udev related, but udev does help show the bug...

thanks,

greg k-h

Re: udev related ? (Was Re: rmmod deadlocks with 2.6.5-rc[2,3])

[Takashi Iwai]

At Tue, 30 Mar 2004 15:42:51 -0800,
Greg KH wrote:
> 
> On Tue, Mar 30, 2004 at 07:49:11PM -0300, Fr?d?ric L. W. Meunier wrote:
> > OK, I gave 2.6.3-rc3-mm1 a try and could reproduce it.
> 
> Yes, the bug is still there.  It is being activly worked on as I type
> this...
> 
> And no, it's not udev related, but udev does help show the bug...

i've heard that the recent sysfs change since 2.6.5-rc triggered rmmod
hangup of ALSA modules.


Takashi