Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Taylor Carpenter]

[-- Attachment #0: Type: message/rfc822, Size: 1231 bytes --]


kernel: 2.4.7  (using devfs)
devfsd: 1.3.11


[12]# insmod floppy
Using /lib/modules/2.4.7athlon/kernel/drivers/block/floppy.o
inserting floppy driver for 2.4.7athlon
Floppy drive(s): fd0 is 1.44M
FDC 0 is a post-1991 82077

[13]# ls -l /dev/fd0
Unable to handle kernel NULL pointer dereference at virtual address 00000000
 printing eip:
c013c4c4
*pde = 00000000
Oops: 0000
CPU:    0
EIP:    0010:[<c013c4c4>]
EFLAGS: 00010246
eax: 00000000   ebx: 00001000   ecx: ffffffff   edx: 00000000
esi: 00000000   edi: 00000000   ebp: dd439fa4   esp: dd439f5c
ds: 0018   es: 0018   ss: 0018
Process ls (pid: 367, stackpage=dd439000)
Stack: df693ec0 00000000 00001000 c015469b df693ec0 bfffec54 00001000 00000000
       df092840 c01376d2 df693ec0 bfffec54 00001000 df092840 dd438000 bffffead
       00000003 bffffc7c df693ec0 c188e4c0 00003000 dd438000 08057000 00000008
Call Trace: [<c015469b>] [<c01376d2>] [<c0106c2b>]

Code: f2 ae f7 d1 49 89 ce 39 de 0f 47 f3 56 52 8b 44 24 1c 50 e8
Segmentation fault


Please CC replies to taylorcc@codecafe.com as I am not on the list.  Thanks.

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Richard Gooch]

Taylor Carpenter writes:
> Any access to /dev/fd0 gives a kernel oops.  Have not accessed the floppy in a
> while, but noticed the complaint when booting and after looking further.
> 
> kernel: 2.4.7  (using devfs)
> devfsd: 1.3.11

If you think the problem may be devfs-related, a good test is to try
again with CONFIG_DEVFS_FS=n.

However, I run devfs, and I don't see this problem.

> [12]# insmod floppy
> Using /lib/modules/2.4.7athlon/kernel/drivers/block/floppy.o
> inserting floppy driver for 2.4.7athlon
> Floppy drive(s): fd0 is 1.44M
> FDC 0 is a post-1991 82077
> 
> [13]# ls -l /dev/fd0
> Unable to handle kernel NULL pointer dereference at virtual address 00000000
>  printing eip:
> c013c4c4
> *pde = 00000000
> Oops: 0000
> CPU:    0
> EIP:    0010:[<c013c4c4>]
> EFLAGS: 00010246
> eax: 00000000   ebx: 00001000   ecx: ffffffff   edx: 00000000
> esi: 00000000   edi: 00000000   ebp: dd439fa4   esp: dd439f5c
> ds: 0018   es: 0018   ss: 0018
> Process ls (pid: 367, stackpage=dd439000)
> Stack: df693ec0 00000000 00001000 c015469b df693ec0 bfffec54 00001000 00000000
>        df092840 c01376d2 df693ec0 bfffec54 00001000 df092840 dd438000 bffffead
>        00000003 bffffc7c df693ec0 c188e4c0 00003000 dd438000 08057000 00000008
> Call Trace: [<c015469b>] [<c01376d2>] [<c0106c2b>]
> 
> Code: f2 ae f7 d1 49 89 ce 39 de 0f 47 f3 56 52 8b 44 24 1c 50 e8
> Segmentation fault

You need to run this through ksymoops before anyone can help you.

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Taylor Carpenter]

On Thu, Aug 16, 2001 at 10:19:07PM -0600, Richard Gooch wrote:
> If you think the problem may be devfs-related, a good test is to try
> again with CONFIG_DEVFS_FS=n.

I tried kernel 2.4.8 and also had an oops.  An earlier kernel w/out devfs did
not cause an oops.  I plan on making a non-devfs 2.4.8 kernel to see if the
oops does not happen.

> However, I run devfs, and I don't see this problem.

I do not know if the devfs info, in the oops (ksymoops output) data below
indicates a problem w/devfs or not.  Maybe you can tell?

> You need to run this through ksymoops before anyone can help you.

Here is the output from ksymoops

-----
>>EIP; c013c4c4 <vfs_readlink+24/60>   <=====
Trace; c015469b <devfs_unregister_blkdev+12eb/1960>
Trace; c01376d2 <cdput+5e2/880>
Trace; c0106c2b <__up_wakeup+108b/24a0>
Code;  c013c4c4 <vfs_readlink+24/60>
00000000 <_EIP>:
Code;  c013c4c4 <vfs_readlink+24/60>   <=====
   0:   f2 ae                     repnz scas %es:(%edi),%al   <=====
Code;  c013c4c6 <vfs_readlink+26/60>
   2:   f7 d1                     not    %ecx
Code;  c013c4c8 <vfs_readlink+28/60>
   4:   49                        dec    %ecx
Code;  c013c4c9 <vfs_readlink+29/60>
   5:   89 ce                     mov    %ecx,%esi
Code;  c013c4cb <vfs_readlink+2b/60>
   7:   39 de                     cmp    %ebx,%esi
Code;  c013c4cd <vfs_readlink+2d/60>
   9:   0f 47 f3                  cmova  %ebx,%esi
Code;  c013c4d0 <vfs_readlink+30/60>
   c:   56                        push   %esi
Code;  c013c4d1 <vfs_readlink+31/60>
   d:   52                        push   %edx
Code;  c013c4d2 <vfs_readlink+32/60>
   e:   8b 44 24 1c               mov    0x1c(%esp,1),%eax
Code;  c013c4d6 <vfs_readlink+36/60>
  12:   50                        push   %eax
Code;  c013c4d7 <vfs_readlink+37/60>
  13:   e8 00 00 00 00            call   18 <_EIP+0x18> c013c4dc <vfs_readlink+3c/60>
-----

After watching what happens during the boot process I started turning off
things that might be touching the floppy device (at boot), such as autofs.
What finally stopped the Oops was to comment out the floppy entry in
/etc/fstab:

	/dev/fd0 /floppy auto defaults,user,noauto 0 0

I can then load the floppy module after boot, and successfully access the
floppy device w/o any problems.  The oops happened when the mountall script
(Debian testing) was running:

	mount -avt nonfs,noproc,nosmbfs

Devfsd is started way before mountall is ran so I do not know why it is
causing problems.

Please CC replies to taylorcc@codecafe.com as I am not on the list.  Thanks.

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Steve Hill]

On Tue, 21 Aug 2001, Taylor Carpenter wrote:

> On Thu, Aug 16, 2001 at 10:19:07PM -0600, Richard Gooch wrote:
> > If you think the problem may be devfs-related, a good test is to try
> > again with CONFIG_DEVFS_FS=n.
> 
> I tried kernel 2.4.8 and also had an oops.  An earlier kernel w/out devfs did
> not cause an oops.  I plan on making a non-devfs 2.4.8 kernel to see if the
> oops does not happen.

I've noticed kernels 2.4.7, 2.4.8 and 2.4.9 can oops when modprobing
floppy.o under certain circumstances (specifically I've noticed it when
rc.sysinit does "/sbin/pam_console_apply -r",which appears to cause the
floppy driver to be modprobed).  There doesn't seem to be a problem if the
floppydriver is compiled into the kernel instead of being modular.

NOTE: I'm testing these kernels on Cobalt kit, which doesn't have a floppy
drive (and does some very odd things at the hardware/firmware side
occasionally :).  So I can't vouch for how this will affect
"normal" hardware.

-- 

- Steve Hill
System Administrator         Email: steve@navaho.co.uk
Navaho Technologies Ltd.       Tel: +44-870-7034015

        ... Alcohol and calculus don't mix - Don't drink and derive! ...

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Alan Cox]

> I've noticed kernels 2.4.7, 2.4.8 and 2.4.9 can oops when modprobing
> floppy.o under certain circumstances (specifically I've noticed it when

floppy.o crashes the box or prints bad things if there is no floppy 
controller. I've not yet had time to investigate

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Steve Hill]

On Wed, 22 Aug 2001, Alan Cox wrote:

> > I've noticed kernels 2.4.7, 2.4.8 and 2.4.9 can oops when modprobing
> > floppy.o under certain circumstances (specifically I've noticed it when
> 
> floppy.o crashes the box or prints bad things if there is no floppy 
> controller. I've not yet had time to investigate

Ok - it's no big problem, I just don't compile it as a module now (the
kernel is used on both boxes with floppy and without, so it needs to be
there, but I have no problem with compiling it into the kernel - it
doesn't seem to break, just says "no floppy controller present" on the
floppy-less boxes.)

-- 

- Steve Hill
System Administrator         Email: steve@navaho.co.uk
Navaho Technologies Ltd.       Tel: +44-870-7034015

        ... Alcohol and calculus don't mix - Don't drink and derive! ...

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Alan Cox]

> there, but I have no problem with compiling it into the kernel - it
> doesn't seem to break, just says "no floppy controller present" on the
> floppy-less boxes.)

Timing dependant - sometimes you get no floppy controller, sometimes a crash
and sometimes moans about things still being active

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Richard Gooch]

Taylor Carpenter writes:
> 
> On Thu, Aug 16, 2001 at 10:19:07PM -0600, Richard Gooch wrote:
> > If you think the problem may be devfs-related, a good test is to try
> > again with CONFIG_DEVFS_FS=n.
> 
> I tried kernel 2.4.8 and also had an oops.  An earlier kernel w/out devfs did
> not cause an oops.  I plan on making a non-devfs 2.4.8 kernel to see if the
> oops does not happen.
> 
> > However, I run devfs, and I don't see this problem.
> 
> I do not know if the devfs info, in the oops (ksymoops output) data below
> indicates a problem w/devfs or not.  Maybe you can tell?

Well, the following line suggests a devfs interaction (provided you
are sure that you used the correct System.map:
> Trace; c015469b <devfs_unregister_blkdev+12eb/1960>

Are you unloading the floppy driver at some point? Or using module
autoloading?

> After watching what happens during the boot process I started turning off
> things that might be touching the floppy device (at boot), such as autofs.
> What finally stopped the Oops was to comment out the floppy entry in
> /etc/fstab:
> 
> 	/dev/fd0 /floppy auto defaults,user,noauto 0 0

Interesting. What happens if you use /dev/floppy/0 instead of /dev/fd0
in the /etc/fstab? Still an Oops?

> I can then load the floppy module after boot, and successfully access the
> floppy device w/o any problems.  The oops happened when the mountall script
> (Debian testing) was running:
> 
> 	mount -avt nonfs,noproc,nosmbfs
> 
> Devfsd is started way before mountall is ran so I do not know why it is
> causing problems.

Strange. I tried the following sequence, without problems:
# devfsd /dev
# mount /floppy
# ls -lF /dev/floppy

where /etc/fstab has:
/dev/floppy/0  /floppy  ext2  defaults,noauto,user  0 0

I have nothing in my /etc/devfsd.conf which refers to the floppy. In
addition, I don't have any LOOKUP entries. I did this with 2.4.9 with
devfs-patch-v188.

Could you please try reproducing the problem with the setup and
sequence I used?

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Taylor Carpenter]

On Tue, Aug 21, 2001 at 11:07:06PM -0600, Richard Gooch wrote:
> Taylor Carpenter writes:
> > 
> Well, the following line suggests a devfs interaction (provided you
> are sure that you used the correct System.map:
> > Trace; c015469b <devfs_unregister_blkdev+12eb/1960>

I can check again, but I am fairly certain it was the correct System.map (I
looked before and after when running ksymoops).

> Are you unloading the floppy driver at some point? Or using module
> autoloading?

I am using module autoloading.  It seems that right after the mountall script
is called this happens.


> Interesting. What happens if you use /dev/floppy/0 instead of /dev/fd0
> in the /etc/fstab? Still an Oops?

I will try that...  

I did not get the oops, but I could not access the device until I manually
loaded the floppy module (the /dev/fd0 was created as usual).

After booting w/the floppy module not loaded I can start and stop devfsd and
load and unload the floppy module in any order w/o an oops. I only get the
oops if the module is loaded during boot.

Also when I get the oops during boot, the floppy module is not loaded once the
system is through booting and I login, but /dev/fd* are all there!  With out
the oops /dev/fd* is only there when I load the floppy module (and disappears
as usual when I rmmod the module).


> Strange. I tried the following sequence, without problems:
> # devfsd /dev
> # mount /floppy
> # ls -lF /dev/floppy

I could not mount the floppy until I manually loaded the floppy module!  This
is after editing /etc/fstab as you said below.

> 
> where /etc/fstab has:
> /dev/floppy/0  /floppy  ext2  defaults,noauto,user  0 0
> 
> I have nothing in my /etc/devfsd.conf which refers to the floppy.

I only reference floppy in my perms file.

> In addition, I don't have any LOOKUP entries. I did this with 2.4.9 with
> devfs-patch-v188.

I have many LOOKUP entries.  I did not have any specific to /dev/fd*, but that
was caught by the .* LOOKUP entry.  I commented that out, which is probably
why mdir and other commands do not work w/o manually loading floppy.o.

> Could you please try reproducing the problem with the setup and sequence I
> used?

I believe I followed your steps, and had no oops during boot w/the /etc/fstab
change.  I do not think that made the difference though.  I think telling
devfsd to not autoload floppy is what let me boot w/o oops.  So I think there
still is a problem w/loading the floppy module during boot.

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Richard Gooch]

Taylor Carpenter writes:
> On Tue, Aug 21, 2001 at 11:07:06PM -0600, Richard Gooch wrote:
> > Taylor Carpenter writes:
> > Trace; c015469b <devfs_unregister_blkdev+12eb/1960>
> > Are you unloading the floppy driver at some point? Or using module
> > autoloading?
> 
> I am using module autoloading.  It seems that right after the
> mountall script is called this happens.

You mean the floppy module gets loaded as a result of the mountall
script being run?

> > Interesting. What happens if you use /dev/floppy/0 instead of /dev/fd0
> > in the /etc/fstab? Still an Oops?
> 
> I will try that...  
> 
> I did not get the oops, but I could not access the device until I manually
> loaded the floppy module (the /dev/fd0 was created as usual).

Is the /etc/modules.devfs the one that ships with devfsd?
Specifically, you need the following line:
alias     /dev/floppy		floppy

and of course the MODLOAD action in /etc/devfsd.conf.

> After booting w/the floppy module not loaded I can start and stop
> devfsd and load and unload the floppy module in any order w/o an
> oops. I only get the oops if the module is loaded during boot.
> 
> Also when I get the oops during boot, the floppy module is not
> loaded once the system is through booting and I login, but /dev/fd*
> are all there!  With out the oops /dev/fd* is only there when I load
> the floppy module (and disappears as usual when I rmmod the module).

Do you mean to say that the Oops happens during the boot sequence,
before you can log in?

> > Strange. I tried the following sequence, without problems:
> > # devfsd /dev
> > # mount /floppy
> > # ls -lF /dev/floppy
> 
> I could not mount the floppy until I manually loaded the floppy
> module!  This is after editing /etc/fstab as you said below.

You must have a bogus /etc/devfsd.conf or /etc/modules.devfs file for
module autoloading not to work.

> > where /etc/fstab has:
> > /dev/floppy/0  /floppy  ext2  defaults,noauto,user  0 0
> > 
> > I have nothing in my /etc/devfsd.conf which refers to the floppy.
> 
> I only reference floppy in my perms file.

I don't know what a "perms" file is. I know Debian uses some
convoluted directory tree for devfsd configuration, but frankly, I
don't want to know about that. If you have one of these monstrosities,
please collapse them all into a single /etc/devfsd.conf file a report
based on that.

> > In addition, I don't have any LOOKUP entries. I did this with 2.4.9 with
> > devfs-patch-v188.
> 
> I have many LOOKUP entries.  I did not have any specific to
> /dev/fd*, but that was caught by the .* LOOKUP entry.  I commented
> that out, which is probably why mdir and other commands do not work
> w/o manually loading floppy.o.

Oh! Well, duh!

> > Could you please try reproducing the problem with the setup and sequence I
> > used?
> 
> I believe I followed your steps, and had no oops during boot w/the
> /etc/fstab change.  I do not think that made the difference though.
> I think telling devfsd to not autoload floppy is what let me boot
> w/o oops.  So I think there still is a problem w/loading the floppy
> module during boot.

Hm. Please try a virgin 2.4.9 kernel with CONFIG_DEVFS_FS=n and try
again. I have this feeling that I'm chasing someone else's bug...

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Taylor Carpenter]

On Wed, Aug 22, 2001 at 11:05:54PM -0600, Richard Gooch wrote:
> Taylor Carpenter writes:
> >
> You mean the floppy module gets loaded as a result of the mountall
> script being run?

As far as I can tell, when it runs mount -avt, the floppy module
is loaded if there is an entry in /etc/fstab for /dev/fd0 (even w/the noauto
option).
 
> Is the /etc/modules.devfs the one that ships with devfsd?
> Specifically, you need the following line:
> alias     /dev/floppy		floppy

I do not have a modules.devfs.  I do have a /etc/modutils/devfsd, but it does
not have an alias entry for the floppy.

I added the entry, and it now shows up in modules.conf.

> and of course the MODLOAD action in /etc/devfsd.conf.

I did not have anything specific to /dev/fd* in devfsd.conf, but I did have a
.* entry to load it.  I commented that out, after your other email.

I added /dev/fd0, and /dev/floppy MODLOAD but it is still not autoloading.
 
> Do you mean to say that the Oops happens during the boot sequence,
> before you can log in?

Yes, the oops happens during boot when it runs the mountall script, and then
right after when something else tries to access the device.  I did not mention
before, but devfsd will not run after I get this oops.  Or I should say it
dies, and then I can not run it again after logging in.  I get an oops
everytime I access the floppy device after this, but I only get an oops if I
get one during boot.

> You must have a bogus /etc/devfsd.conf or /etc/modules.devfs file for
> module autoloading not to work.

I had commented out the .* MODLOAD, it worked before that.  I uncommented the
.* and it still does not work, something else must have changed.

> I don't know what a "perms" file is. I know Debian uses some
> convoluted directory tree for devfsd configuration, but frankly, I
> don't want to know about that. If you have one of these monstrosities,
> please collapse them all into a single /etc/devfsd.conf file a report
> based on that.

OK.  I have nothing but REGISTER PERMISSIONS for floppy in devfsd.conf.

> > I have many LOOKUP entries.  I did not have any specific to
> > /dev/fd*, but that was caught by the .* LOOKUP entry.  I commented
> > that out, which is probably why mdir and other commands do not work
> > w/o manually loading floppy.o.
> 
> Oh! Well, duh!

I uncommented .* LOOKUP and it still did not autoload.

> Hm. Please try a virgin 2.4.9 kernel with CONFIG_DEVFS_FS=n and try
> again. I have this feeling that I'm chasing someone else's bug...

OK, done.  I did not get the oops with DEVFS turned off (in 2.4.9).

Re: Oops when accessing /dev/fd0 (kernel 2.4.7 and devfsd 1.3.11)

[Richard Gooch]

Taylor Carpenter writes:
> On Wed, Aug 22, 2001 at 11:05:54PM -0600, Richard Gooch wrote:
> > Taylor Carpenter writes:
> > >
> > You mean the floppy module gets loaded as a result of the mountall
> > script being run?
> 
> As far as I can tell, when it runs mount -avt, the floppy module
> is loaded if there is an entry in /etc/fstab for /dev/fd0 (even w/the noauto
> option).
>  
> > Is the /etc/modules.devfs the one that ships with devfsd?
> > Specifically, you need the following line:
> > alias     /dev/floppy		floppy
> 
> I do not have a modules.devfs.  I do have a /etc/modutils/devfsd,
> but it does not have an alias entry for the floppy.

What the hell is /etc/modutils/devfsd? More Debian tricks?

[...]
> > You must have a bogus /etc/devfsd.conf or /etc/modules.devfs file for
> > module autoloading not to work.
> 
> I had commented out the .* MODLOAD, it worked before that.  I uncommented the
> .* and it still does not work, something else must have changed.
> 
> > I don't know what a "perms" file is. I know Debian uses some
> > convoluted directory tree for devfsd configuration, but frankly, I
> > don't want to know about that. If you have one of these monstrosities,
> > please collapse them all into a single /etc/devfsd.conf file a report
> > based on that.
> 
> OK.  I have nothing but REGISTER PERMISSIONS for floppy in devfsd.conf.

Argh! Let me spell it out: I want you to collapse all devfsd
configuration files into a single file, and throw the stuff in
directories away. And I also want you to do the same for modutils
configuration files. I don't want you to edit out stuff that you think
isn't relevant. I want the whole bloody lot so I can see the context.

Once you've stripped down the configuration to something simple, check
if the Oops still happens. If so, decode it with ksymoops and send it.
Hm. And also send the complete output of dmesg. No editing.

> > Hm. Please try a virgin 2.4.9 kernel with CONFIG_DEVFS_FS=n and try
> > again. I have this feeling that I'm chasing someone else's bug...
> 
> OK, done.  I did not get the oops with DEVFS turned off (in 2.4.9).

Interesting. But I need those collapsed files. I have no desire to
figure out the subtle effects that can be generated by Debian's
configuration system.

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca