On 04/03/2007 08:57 AM, Pekka J Enberg wrote:

[ re-including linux-kernel ]

> Does this change the dd case?
> 
> diff --git a/drivers/cdrom/mcdx.c b/drivers/cdrom/mcdx.c
> index f574962..6c613d0 100644
> --- a/drivers/cdrom/mcdx.c
> +++ b/drivers/cdrom/mcdx.c
> @@ -1248,6 +1248,7 @@ #endif
>  	disk->private_data = stuffp;
>  	disk->queue = mcdx_queue;
>  	add_disk(disk);
> +	blk_queue_hardsect_size(mcdx_queue, MCDX_CDBLK);
>  	printk(msg);
>  	return 0;
>  }

No, I'm afraid not. It's still the same effect:

root@5va2:~# dmesg -c >/dev/null
root@5va2:~# strace -o dd.strace dd if=/dev/mcdx0 of=/dev/null
0+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00438956 seconds, 0.0 kB/s
root@5va2:~# dmesg >dd.xtrace

dd.strace again attached. I've dropped the mutex patch for now and now
the xtrace is littered with waitqueue debugging. I'll see if I can
disable that particular debugging trace I guess. dd.xtrace sent privately...

"readcd dev=/dev/mcdx0 f=foo.img" is still an immediate hard lockup also
without the mutex. I believe the "tar" runs has produced something
useful now though:

root@5va2:~# dmesg -c >/dev/null
root@5va2:~# strace -o tar.strace tar cvf foo.tar /mnt/cdrom
tar: Removing leading `/' from member names
/mnt/cdrom/
/mnt/cdrom/dott/
/mnt/cdrom/dott/adlib.ims
/mnt/cdrom/dott/dott.bat
/mnt/cdrom/dott/dotticon.ico
/mnt/cdrom/dott/gmidi.ims
/mnt/cdrom/dott/maniac/
rlogin: connection closed.
rene@7ixe4:~$

When I now switch the monitor to 5va2 directly, there is an oops on the
screen, with a backtrace saying (manual transcription):

Call Trace:
   [<c1047eb6>] drain_array+0x94/0xc3
   [   ...    ] cache_reap+0x55/0xe9
   [   ...    ] run_workqueue+0x8f/0x14d
   [   ...    ] cache_reap+0x0//0xe9
   [   ...    ] worker_thread+0x122/0x14b
   [   ...    ] default_wake_function+0x0/0xc
   [   ...    ] default_wake_function+0x0/0xc
   [   ...    ] worker_thread+0x0/0x14b
   [   ...    ] kthread+0x72/0x97
   [   ...    ] kthread+0x0/0x97
   [   ...    ] kernel_thread_helper+0x7/0x10
   =======================
Code: [ ... ]
EIP: [<c1047787>] free_block+0x58/0xcf SS:ESP [ ... ]

Couple of functions mentioned twice in that backtrace. That looks a
little off no?

"tar.strace" made it to disk, so it's attached as well.

Rene.