Skip to contentSkip to conversation

oops with 2.4.18-pre9-mjc2

DISCUSSION Robert Jameson 9 participants 16 messages Feb 15, 2002 – Feb 16, 2002

lore
Share & tools
lore
#0 lore raw

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Reading Oops report from the terminal

Unable to handle kernel paging request at virtual address 0004004c
dc838114
*pde = 00000000
Oops: 0000
CPU:    0
EIP:    0010:[<dc838114>]    Tainted: P 

Using defaults from ksymoops -t elf32-i386 -a i386
EFLAGS: 00013202
eax: 00040038 ebx: d335d094 ecx: cc83c000 edx: 00040038
esi: cc83c560 edi: d335d000 ebp: d335d0f0 esp: c7223ea4
ds: 0018 es: 0018 ss: 0018
Process jpilot (pid: 23909, stackpage=c7223000)

Stack: dc9472df 00040038 d335d094 d335d000 c9299ec0 00000000 dc845484

d335d094 c9299ec0 c2743000 c859e280 c142e320 c01a2990 c2743000
c9299ec0 c9299ec0 c859e280 c142e320 c5bfb240 00000001 00007053
00003829 00000000 c012b45b Call Trace: [<dc9472df>] [<dc845484>]
[<c01a2990>] [<c012b45b>] [<c011fb50>] [<c011ff72>] [<c01a2fb6>]
[<c0131044>] [<c013014c>] [<c01162aa>] [<c0116834>] [<c011697a>]
[<c0106d13>] Code: 8b 42 14 85 c0 74 1b 8b 80 bc 00 00 00 85 c0 74 11 8b 40 18

Quoted 1 line
Quoted 1 line

EIP; dc838114 <[usbcore]usb_unlink_urb+8/30> <=====

Trace; dc9472de <[visor]visor_close+13a/168>
Trace; dc845484 <[usbserial]serial_close+a0/b0> Trace; c01a2990 <release_dev+240/4fc>
Trace; c012b45a <free_page_and_swap_cache+32/38> Trace; c011fb50 <__free_pte+40/48>
Trace; c011ff72 <do_zap_page_range+18e/238>
Trace; c01a2fb6 <tty_release+a/10>
Trace; c0131044 <fput+4c/d0>
Trace; c013014c <filp_close+5c/64>
Trace; c01162aa <put_files_struct+4e/b4>
Trace; c0116834 <do_exit+a8/1c8>
Trace; c011697a <sys_exit+e/10>
Trace; c0106d12 <system_call+32/38>
Code; dc838114 <[usbcore]usb_unlink_urb+8/30> 00000000 <_EIP>:
Code; dc838114 <[usbcore]usb_unlink_urb+8/30> <=====
0: 8b 42 14 mov 0x14(%edx),%eax <=====
Code; dc838116 <[usbcore]usb_unlink_urb+a/30>
3: 85 c0 test %eax,%eax
Code; dc838118 <[usbcore]usb_unlink_urb+c/30>
5: 74 1b je 22 <_EIP+0x22> dc838136
<[usbcore]usb_unlink_urb+2a/30> Code; dc83811a
<[usbcore]usb_unlink_urb+e/30> 7: 8b 80 bc 00 00 00 mov
0xbc(%eax),%eax Code; dc838120 <[usbcore]usb_unlink_urb+14/30>
d: 85 c0 test %eax,%eax
Code; dc838122 <[usbcore]usb_unlink_urb+16/30>
f: 74 11 je 22 <_EIP+0x22> dc838136
<[usbcore]usb_unlink_urb+2a/30> Code; dc838124
<[usbcore]usb_unlink_urb+18/30> 11: 8b 40 18 mov
0x18(%eax),%eax

This error occurs while hotsyncing my handspring visor pda witth pilot-link/jpilot

Signature · 4 lines
Robert Jameson                  http://rj.open-net.org
C2 Village at Wexford Hwy 278,  Tel: +1 (843) 757 9428
Hilton Head Isl, SC             Cel: +1 (843) 298 0957 
US, 29928.                      mailto:[email protected]
#1 lore raw
Quoted 1 line · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

This sounds like you might need bits from -gregkh as well.

Cheers,
Bill

#2 lore raw
In reply to Robert Jameson
Quoted 1 line · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Ouch. Does stock 2.4.18-pre9 do this?

------
Michael Cohen
OhDarn.net

#3 lore raw
In reply to Robert Jameson
Quoted 1 line · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Do you see this on device close? It looks like there may be a race between device closer -> usb release.

Can you reproduce it without the binary module you are loading?

Robert Love

#4 lore raw

It's appears right after my PDA finishes syncing, so im guessing, its during a device close. To answer alans question im using nVidias kernel driver, therefor i tainted the kernel (tm) (c).

Quoted 10 lines · On 15 Feb 2002 08:37:52 -0500 Robert Love <[email protected]> wrote:
Quoted 2 lines · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Do you see this on device close? It looks like there may be a race between device closer -> usb release.

Can you reproduce it without the binary module you are loading?

Robert Love

Signature · 4 lines
Robert Jameson                  http://rj.open-net.org
C2 Village at Wexford Hwy 278,  Tel: +1 (843) 757 9428
Hilton Head Isl, SC             Cel: +1 (843) 298 0957 
US, 29928.                      mailto:[email protected]
#5 lore raw
Quoted 3 lines

It's appears right after my PDA finishes syncing, so im guessing, its during a device close. To answer alans question im using nVidias kernel driver, therefor i tainted the kernel (tm) (c).

Please take your bug report to Nvidia. You'll find the binary module needs recompiling for pre-empt. Have fun with them 8)

#6 lore raw
Quoted 2 lines · Alan Cox

Please take your bug report to Nvidia. You'll find the binary module needs recompiling for pre-empt. Have fun with them 8)

According to his config, preempt-kernel wasn't enabled.

We don't tend to see problems with preempt-kernel + evil-closed-nvidia driver (for whatever odd reason), anyway.

Robert Love

#7 lore raw
In reply to Robert Jameson
Quoted 5 lines

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Reading Oops report from the terminal

CPU:    0
EIP:    0010:[<dc838114>]    Tainted: P 
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

What strange modules do you have loaded ?

If its a binary only one I'd like to know. If its just the base kernel I'd appreciate an lsmod so I can find which module is missing a license tag

#8 lore raw
In reply to Robert Jameson
Quoted 4 lines · Robert Jameson

It's appears right after my PDA finishes syncing, so im guessing, its during a device close. To answer alans question im using nVidias kernel driver, therefor i tainted the kernel (tm) (c).

you're using a binary only kernel driver AND a preempt kernel ? brave. very brave.

preempt works on the assumption that it can change the content of inline functions and such....

#9 lore raw
In reply to Robert Jameson
Quoted 1 line · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Known problem, sorry. I can't duplicate this myself to try to fix this. Other people have reported workarounds by using a different host controller driver, or running a SMP kernel on a UP machine.

Patches gladly accepted :)

thanks,

greg k-h

#10 lore raw

Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

In reply to Robert Jameson
Quoted 6 lines · Robert Love
Quoted 1 line · Robert Jameson

I have been seeing this oops from 2.4.16 -> 2.4.18-pre9, so here we go!

Do you see this on device close? It looks like there may be a race between device closer -> usb release.

I don't use USB, and I have had several machines lock up hard while doing medium to heavy IO. I've had this happen with pre9-mjc2 and another patch that just contained pre9-preempt-schedo1 (nyu.dyn.dhs.org:8080/patches/2.4.18-pre9-to-rmap12e-schedO1-rml.patch.bz2)

I'm running 2.4.18-pre9-ac3 now to see if I can reproduce without prempt and O(1).

I have someone else from IRC that has the same issue with prempt+O(1) against vanilla 2.4.17. He should be sending you a bug report soon.

BTW, all machines ran the same kernel compiled for SMP, but some machines were UP.

Has anyone else seen this?

Mike

#11 lore raw

Re: Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

Quoted 4 lines · Mike Fedyk

I don't use USB, and I have had several machines lock up hard while doing medium to heavy IO. I've had this happen with pre9-mjc2 and another patch that just contained pre9-preempt-schedo1 (nyu.dyn.dhs.org:8080/patches/2.4.18-pre9-to-rmap12e-schedO1-rml.patch.bz2)

The -mjc and similar patches make debugging a bit, uh, hard ;)

Quoted 2 lines

I'm running 2.4.18-pre9-ac3 now to see if I can reproduce without prempt and O(1).

If you can't reproduce it, I'd like to see if you can reproduce it _only_ with preempt. Also, if it happens on stock pre9 (no -ac) would be of interest, since that doesn't have Andre's IDE patch.

Quoted 2 lines

I have someone else from IRC that has the same issue with prempt+O(1) against vanilla 2.4.17. He should be sending you a bug report soon.

Now this would be of interest, thanks.

Quoted 4 lines

BTW, all machines ran the same kernel compiled for SMP, but some machines were UP.

Has anyone else seen this?

Robert Love

#12 lore raw

Re: Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

In reply to Mike Fedyk
Quoted 7 lines · Mike Fedyk

I don't use USB, and I have had several machines lock up hard while doing medium to heavy IO. I've had this happen with pre9-mjc2 and another patch that just contained pre9-preempt-schedo1 (nyu.dyn.dhs.org:8080/patches/2.4.18-pre9-to-rmap12e-schedO1-rml.patch.bz2)

I'm running 2.4.18-pre9-ac3 now to see if I can reproduce without prempt and O(1).

I run the same configuration here but UP, with quite a bit of I/O, box NFS exports a large build directory to 2 other boxes plus usually has 2 builds going locally as well as being used as a storage area for creating isos/burning cds. VM load is medium for a 512mem/512swap box, ~140 processes and ~150megs into swap.

Cheers,
Zwane Mwaikambo

#13 lore raw

Re: Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

In reply to Robert Jameson
Quoted 9 lines · Robert Love
Quoted 4 lines · Mike Fedyk

I don't use USB, and I have had several machines lock up hard while doing medium to heavy IO. I've had this happen with pre9-mjc2 and another patch that just contained pre9-preempt-schedo1 (nyu.dyn.dhs.org:8080/patches/2.4.18-pre9-to-rmap12e-schedO1-rml.patch.bz2)

The -mjc and similar patches make debugging a bit, uh, hard ;)

Yep, I understand. When I was patching in rmap12f I had to manually merge the little bit into mm/bootmem.c and the offset was several hundred lines. Then I realized just how much WLI's bootmem patch changes.

Quoted 7 lines
Quoted 2 lines

I'm running 2.4.18-pre9-ac3 now to see if I can reproduce without prempt and O(1).

If you can't reproduce it, I'd like to see if you can reproduce it _only_ with preempt. Also, if it happens on stock pre9 (no -ac) would be of interest, since that doesn't have Andre's IDE patch.

Actually, I'm going to recompile -mjc2 without lock breaking to see if that helps. Then try without prempt altogether. If either of those two fix the problem, I'll see if I can reproduce against the latest kernel from marcello and your latest patch and merge myself. Heh, I want to keep testing -mjc. There are so many nice things in there. ;)

Quoted 5 lines
Quoted 2 lines

I have someone else from IRC that has the same issue with prempt+O(1) against vanilla 2.4.17. He should be sending you a bug report soon.

Now this would be of interest, thanks.

I asked him to cc me so that I may be able to help too...

Quoted 1 line

Robert Love

Mike

#14 lore raw

Re: Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

In reply to Robert Jameson
Quoted 3 lines · Mike Fedyk

Yep, I understand. When I was patching in rmap12f I had to manually merge the little bit into mm/bootmem.c and the offset was several hundred lines. Then I realized just how much WLI's bootmem patch changes.

It's a rewrite. Of course it changes the whole file. Lucky for you it interacts with nothing else. I seem to remember this conflict being somewhat trivial to resolve though.

Cheers,
Bill

#15 lore raw

Re: Hard lockup with 2.4.18-pre9 + preempt + lock break + O1k[23] + rmap

In reply to Robert Jameson
Quoted 8 lines · William Lee Irwin III
Quoted 3 lines · Mike Fedyk

Yep, I understand. When I was patching in rmap12f I had to manually merge the little bit into mm/bootmem.c and the offset was several hundred lines. Then I realized just how much WLI's bootmem patch changes.

It's a rewrite. Of course it changes the whole file. Lucky for you it interacts with nothing else. I seem to remember this conflict being somewhat trivial to resolve though.

Yes, it was quite easy to hand merge/fix it up when I added in rmap12f (was 12e... so not much to change.)

Mike