I've been having a really annoying problem for quite some time
now, that has completly stumped me. The problem is that on one
of my hard drives, when connected to my PCI expansion IDE
contrtoller, and I boot a kernel that uses DMA, I get massive
filesystem corruption & total data loss.

The affected drive is a Western Digital WD205AA-00BAA0 20GB
UDMA66. It is attached to a Promise Ultra133TX2 (PDC20269) PCI
expansion IDE controller. At this point I don't even know which
of the two (if either) actually is the source of the problem, because
the hard drive works fine when attached to my motherboards IDE
controller, and the Ultra133TX2 controller also has two other hard
drives attached that work perfectly. Another detail that adds to
my confusion is that under M$ Windows XP, the hard drive works
fine when attached to the Ultra133TX2.

For quite some time I've been using a work-around that allowed
me to use the drive & controller together. The work-around
indicates that DMA may be related to the problem, although again
there is some very confusing stuff here. My work-around was to
disable DMA in the kernel, by turning off CONFIG_IDEDMA_AUTO
& CONFIG_IDEDMA_PCI_AUTO in make menuconfig, and then
enabling DMA with 'hdparm -d1' in my bootscripts. When I was
doing this, everything worked fine. But since I couldn't leave well
enough alone, I was trying to figure out why DMA works when
enabled with hdparm, but not through the kernel, and as I was
doing so, I lost everything on the hard drive (fortunatly backed
up), am no closer to knowing what the root problem is, and worst
of all haven't been able to restore my work-around configuration.

I've tried kernels 2.4.18 through 2.4.22, and 2.6.0-test4, and all
of them behave the same way.

[Edit]In the process of troubleshooting this, I've contacted both
Promise and Western Digital. So far I haven't got much info from
either of them.[/Edit]

Any help on this would be hugely appreciated,
Beolach

System configuration:
Distribution: Slackware 9.0 (Kernel 2.4.22)
CPU: AMD Athlon XP 2000+
MoBo: MSI KT3 Ultra
MoBo Chipset: VIA KT333 & VT8233A
Memory: 768 MB DDR
IDE Configuration:
Promise Ultra133TX2 (PDC20269) PCI Expansion IDE controller
IDE0
hda - Maxtor 6L060J3 60GB UDMA133 (MS Windows partitions)
hdb - unused
IDE1
hdc - Western Digital WD205AA-00BAA0 20GB UDMA66
hdc1 - ext3 root filesystem
hdc2 - swap partition
hdd - Maxtor 6Y120P0 120GB UDMA133 (backup of hdc)
MoBo Onboard Controller (VIA VT8233A)
IDE2
hde - Lite-On LTR-40125S, ATAPI CD-RW
hdf - Toshiba SD-M1502, ATAPI DVD-ROM
IDE3
hdg - Imation LS-120 SuperDisk, ATAPI Floppy
hdh - unused

ok, i heard of this from the mandrake-linux mail list that i suscribe rto...

western digital does not use the full specs for something or another (am i vague enough?) making them suseptible to file system corruption. sorry i can't be more specific but a google search should be more specific but i know i'm right about western digital not being a hard drive for linux users. BTW, winblows incorporates a work-around for western digital hard drives.

-cheers

Thanks, Iggy.

I actually have read many posts on mailing list from searching
through google. Some of them do agree with what you said, that
WD drives have problems with DMA under Linux. What's really
confusing me about this is that when I enabled DMA using
'hdparm -d1' I would get DMA performance with no problems, but
when I set CONFIG_IDEDMA_AUTO & CONFIG_IDEDMA_PCI_AUTO
in the kernel config, I would get massive filesystem corruption.
What's the difference between enabling DMA in the kernel, and
enabling DMA with hdparm?

Beolach

good question. not being a specialist in kernel subjects (i.e take this with a grain of salt :^) i'd say that they are doing the same thing in different ways.

--cheers

At this point, I do think it is most likely a problem with the hard
drive. But along with DMA working with hdparm, the other thing
that makes me think it may be something else, is that the WD
hard drive has no problems (either with DMA via the kernel or
using hdparm) when I move it to my motherboards IDE controller.

Unfortunatly this isn't a very good solution for me, because I have
several ATAPI devices normally conected to the MoBo IDE, that I
boot off of occasionaly (bootable CDs, and a ATAPI floppy drive),
but the Ultra133TX2 controller doesn't work well with bootable
ATAPI devices, so I can't just swap them with the WD hard drive.

And at this point I'm out of work and strapped for cash, so I can't
just go out and buy a replacement hard drive (and my brother is
nagging me to return the one I borrowed from him). So if I can
get this drive working nicely, I will be a very happy camper.

Beolach

You're going to be mad at me. But, here goes. Maxtor drives on one IDE #, western Digital on another. WD doesn't play well with others except WD (and even some of those don't get along.)

I had the same problem with WD + Samsung. Drove me nuts. Windows worked O.K. but major corruption in linux. On a CMD680 based controller I had problems even with them on different IDE#'s.

One thing though, most of the newer MB's have IDE controllers which do handle the timing difficulties involving two different speed devices as master and slave--not like the old days.

Both of your Maxstors are 133's so make them Master/slave on one of your IDE #'s on your Promise Card. Depending on performance needs, and such, you may have to move partitions around.

Another thing you might want to check up on is whether any of your frequently accessed hardware is sharing IRQs. I read somewhere that this is A Very Bad Thing with WD drives.

Thanks, Eqwatz.

I've actually seen some info on that elsewhere; and I've also
heard that on some IDE controllers each channel can only operate
at one speed, so a UDMA133 and UDMA66 drive on the same
channel would both operate at UDMA66.

But the two different brand drives on the same channel isn't
whats causing my problem. The Maxtor 120GB UDMA133 drive
actually belongs to my brother. I borrowed it from him so I could
have a place to backup everything from the WD drive before trying
to troubleshoot it.

Of course, there's really no reason for it to be on the same
channel as the WD drive, so I will move it to the other channel as
I continue troubleshooting.

I haven't looked into which IRQ's are used by what. Will check it
out & post my findings.

Thanks,
Beolach