Some USB key cant be used on 2.6.15/16 : avoid CONFIG_ACORN_PARTITION_CUMANA ?
I've got a problem (Debian unstable) with linux kernel packages
linux-image-2.6.{15,16}-1-686 : I can't use my USB key.
In the syslog, I found this
SCSI device sda: 2007040 512-byte hdwr sectors (1028 MB)
sda: Write Protect is off
sda: Mode Sense: 23 00 00 00
sda: assuming drive cache: write through
SCSI device sda: 2007040 512-byte hdwr sectors (1028 MB)
sda: Write Protect is off
sda: Mode Sense: 23 00 00 00
sda: assuming drive cache: write through
sda: [CUMANA/ADFS] sda1<5>sd 0:0:0:0: Attached scsi removable disk sda
Buffer I/O error on device sda1, logical block 508378384
Buffer I/O error on device sda1, logical block 508378384
Buffer I/O error on device sda1, logical block 508378412
last message repeated 5 times
I think the problem comes from the partition detection code ([CUMANA/ADFS]),
because the USB key is using simple fat partition. Moreover, the problem
disappeared automagically on another box after I recompiled the kernel
with only options I needed. Nevertheless, the problem is still there on
the box I'm using right now.
Google showed me this interesting page [1] from january 2005 about
a kernel issue that seems to be the culprit:
"Disable CONFIG_ACORN_PARTITION_CUMANA unless you
actually have an Acorn and Cumana partitions."
The standard debian kernels are affected:
# grep CUMANA /boot/config-2.6.{15,16}-1-686
/boot/config-2.6.15-1-686:CONFIG_ACORN_PARTITION_CUMANA=y
/boot/config-2.6.16-1-686:CONFIG_ACORN_PARTITION_CUMANA=y
This can explain why the problem disappeared on the other box after I recompiled
the kernel with custom options. I'll recompile the kernel myself to see if I really
found the culprit, but I'm confident it will work.
As I'm not the only one to have that problem and the vast majority of users
wont recompile their kernel (joe user will simply say "Linux sucks" and go
back to windows/mac), I'm asking myself if the Debian kernels should use
that config option : as it's not a module, it can't be easily disabled.
Is that config option really mandatory ? Should I file a bug somewhere in the BTS ?
Or is it an "upstream", kernel bug ? What's the best way to fix it ?
Christophe
[1] Re: [Bugme-new] [Bug 4054] Linux partition table reading
http://www.ussg.iu.edu/hypermail/linux/kernel/0501.2/0543.html
PS I'll recompile my kernel in the next few days (it's my box at work)
Reply to: