kernel oops when burning CDs

kernel oops when burning CDs

[Rudo Thomas]

Hello everybody!

I get an ooops and immediate kernel panic when I break (CTRL-C) cdrecord. I 
can reproduce it anytime. I use 2.4.5-ac series. Obviously, Linus' 2.4.5 is 
fine.

I know, I know. I was supposed to make a serios oops report, BUT I wasn't 
able to get the serial console running. (I've read some people couldn't get 
it to run either.)

Can anyone help me with any of the problems?

Rudo Thomas.

Re: kernel oops when burning CDs

[Alan Cox]

> I get an ooops and immediate kernel panic when I break (CTRL-C) cdrecord. I 
> can reproduce it anytime. I use 2.4.5-ac series. Obviously, Linus' 2.4.5 is 
> fine.
> I know, I know. I was supposed to make a serios oops report, BUT I wasn't 

Write down the EIP and the call trace then look them up in System.map. Also
include the hardware details. The -ac tree has a newer version of the scsi
generic code. It fixes some oopses but in your case it apparently added a new
failure case

Re: kernel oops when burning CDs

[Tachino Nobuhiro]

Hello,

At Mon, 4 Jun 2001 22:43:30 +0100 (BST),
Alan Cox wrote:
> 
> > I get an ooops and immediate kernel panic when I break (CTRL-C) cdrecord. I 
> > can reproduce it anytime. I use 2.4.5-ac series. Obviously, Linus' 2.4.5 is 
> > fine.
> > I know, I know. I was supposed to make a serios oops report, BUT I wasn't 
> 
> Write down the EIP and the call trace then look them up in System.map. Also
> include the hardware details. The -ac tree has a newer version of the scsi
> generic code. It fixes some oopses but in your case it apparently added a new
> failure case

  Oops occures in SG driver. This patch fixes the problem.



diff -r -u linux.org/drivers/scsi/sg.c linux/drivers/scsi/sg.c
--- linux.org/drivers/scsi/sg.c	Fri Jun  1 10:10:22 2001
+++ linux/drivers/scsi/sg.c	Fri Jun  1 17:08:00 2001
@@ -1114,7 +1114,8 @@
             sg_remove_sfp(sdp, sfp);
 	    sfp = NULL;
         }
-	__MOD_DEC_USE_COUNT(sg_template.module);
+	if (sg_template.module)
+		__MOD_DEC_USE_COUNT(sg_template.module);
 	if (sdp->device->host->hostt->module)
 	    __MOD_DEC_USE_COUNT(sdp->device->host->hostt->module);
     }
@@ -1311,7 +1312,8 @@
 			sg_finish_rem_req(srp);
 		}
 		if (sfp->closed) {
-		    __MOD_DEC_USE_COUNT(sg_template.module);
+		    if (sg_template.module)
+			__MOD_DEC_USE_COUNT(sg_template.module);
 		    if (sdp->device->host->hostt->module)
 			__MOD_DEC_USE_COUNT(sdp->device->host->hostt->module);
 		    __sg_remove_sfp(sdp, sfp);
@@ -2207,7 +2209,8 @@
     else {
         sfp->closed = 1; /* flag dirty state on this fd */
 	/* MOD_INC's to inhibit unloading sg and associated adapter driver */
-	__MOD_INC_USE_COUNT(sg_template.module);
+	if (sg_template.module)
+	    __MOD_INC_USE_COUNT(sg_template.module);
 	 if (sdp->device->host->hostt->module)
 	    __MOD_INC_USE_COUNT(sdp->device->host->hostt->module);
         SCSI_LOG_TIMEOUT(1, printk(

Re: kernel oops when burning CDs

[Rudo Thomas]

> > > I get an ooops and immediate kernel panic when I break (CTRL-C)
> > > cdrecord. I can reproduce on 2.4.5-ac series, Linus' 2.4.5 is fine.
> >
> > ... Also include the hardware details. The -ac tree has a newer version of
> > the scsi generic code.
>
>   Oops occures in SG driver. This patch fixes the problem.

Indeed, the patch seems to have solved the problem. Thanks.

The hardware used was a HP 9100 through ide-scsi on a VIA MVP3 based MB. This 
is probably irrelevant, but just in case...

Rudo Thomas.

ps: Here's what I got to:

>>EIP; c01abe88 <sg_cmd_done_bh+1cc/274>   <=====
Trace; c01a3396 <scsi_old_done+56a/57c>
Trace; c01a6f28 <idescsi_end_request+1f8/240>
Trace; c01a6ff3 <idescsi_pc_intr+83/234>
Trace; c0195a17 <ide_intr+f7/14c>
Trace; c01a6f70 <idescsi_pc_intr+0/234>
Trace; c010798c <simd_math_error+54/c8>
Trace; c0107eee <do_IRQ+6e/b0>
Trace; 0c019efe Before first symbol
Code;  c01abe88 <sg_cmd_done_bh+1cc/274>
00000000 <_EIP>:
Code;  c01abe88 <sg_cmd_done_bh+1cc/274>   <=====
   0:   ff 48 10                  decl   0x10(%eax)   <=====
Code;  c01abe8b <sg_cmd_done_bh+1cf/274>
   3:   a1 ac f3 22 c0            mov    0xc022f3ac,%eax
Code;  c01abe90 <sg_cmd_done_bh+1d4/274>
   8:   80 48 18 08               orb    $0x8,0x18(%eax)
Code;  c01abe94 <sg_cmd_done_bh+1d8/274>
   c:   88 54 24 14               mov    %dl,0x14(%esp,1)
Code;  c01abe98 <sg_cmd_done_bh+1dc/274>
  10:   8b 02                     mov    (%edx),%eax
Code;  c01abe9a <sg_cmd_done_bh+1de/274>
  12:   8b 40 00                  mov    0x0(%eax),%eax

Re: kernel oops when burning CDs

[Douglas Gilbert]

Tachino wrote:
> At Mon, 4 Jun 2001 22:43:30 +0100 (BST),
> Alan Cox wrote:
> > 
> > > I get an ooops and immediate kernel panic when I break (CTRL-C) cdrecord. I 
> > > can reproduce it anytime. I use 2.4.5-ac series. Obviously, Linus' 2.4.5 is 
> > > fine.
> > > I know, I know. I was supposed to make a serios oops report, BUT I wasn't 
> > 
> > Write down the EIP and the call trace then look them up in System.map. Also
> > include the hardware details. The -ac tree has a newer version of the scsi
> > generic code. It fixes some oopses but in your case it apparently added a new
> > failure case
>
>  Oops occures in SG driver. This patch fixes the problem.

[patch snipped. It looks fine]

Tachino,
Thanks for the patch. Alan Cox has incorporated it in lk 2.4.5-ac9 .
The problem only occurs when the sg driver is built in. [I do my
testing with sg built as a module.]

Currently sg version 3.1.18 is in the ac tree while the previous
version (3.1.17) is in the main branch. The code that oopsed is
there to temporarily bump the sg module count to inhibit rmmod
(and/or autoclean) until all SCSI commands on that device are
finished.

Doug Gilbert