[lkml]   [2003]   [Aug]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: FS Corruption with VIA MVP3 + UDMA/DMA
Nothing runs on this one ;-)

WinXP/2003 will die from registry and unrecoverable NTFS filesystem
corruption. Win98 will randomly corrupt driver files eventually leading
to an unbootable system, or worse, a completely corrupted filesystem as
scandisk happily crosslinks all the files (experienced this several
times, just thought it was the hard drives and windows...since the
drives would fail a few months later and since I had past experience
with a Pentium 166 and HX system running Win95 doing this).

Linux fared better, but still would corrupt the filesystem, sometimes
leading to an unusable system say if an important library is moved to
lost+found during fsck. It was much more reliable than any Windows
install and easily repairable. With windows, I had no choice but to
re-install (backing up the registry after every boot worked until NTFS
would eventually die). I lost a few data and help files under linux,
but at this point I backed up all the time anyway (after my first
installation was hopelessly mangled).

I've tried several PCI tweaks with 2.4 which didn't really seem to cure
anything. My powertweak doesn't seem to like the 2.5 series kernels, so
I haven't tried that. Not that it seems to matter, the promise
controllers have much better throughput anyway even with the same modes
and settings in hdparm. I tried all the hdparm combinations of dma
modes and other settings with only a slight decrease in the chance of
corruption and a corresponding dive in throughput. It worked through
2.5.74, but I finally disabled it for everything except my IDE ZIP drive
and stuck in another promise card after concluding that it was just
hopelessly broken.

It would have been nice if 2.4 would just refuse to use DMA, that way
I'd have known about the problem much earlier. I would think with all
the stuff in the kernel about the RZ1000, the problems with the MVP3
would be mentioned as well. As just a typical end user I couldn't
figure out why Linux and reiserfs, which are supposed to be so stable
wouldn't weren't. At this point I'd already run exhaustive memory ,
hard drive bad sector, and CPU tests without any failures so I was
pretty certain it wasn't a hardware issue. Everyone I knew had crashes
with Windows so those didn't surprise me so much.

It's a decent computer for web browsing and let's me gauge the
performance of my business apps. It's a pretty good low-end target
machine now that it doesn't write garbage to my drives.

I just think this should be documented in case someone sets up a
proxy/firewall machine with this configuration. For the majority of
home users, any higher-end machine is probably wasted on such an
application. I setup such a system to share my parents dial-up
connection over a wireless network. Of course, it's using an HX chipset
and P233MMX so it's rock solid, only needing rebooted when the modem
locks up (happened twice since I set it up a year ago). Even though
it's running 2.4.18 and my dad likes to reset it rather than
CTRL-ALT-DEL when the modem locks up, it has yet to corrupt reiserfs.

That's the kind of stability that got me really wondering about my system...

Jamie Lokier wrote:

>insecure wrote:
>>>The VP2/97 also had severe problems with DMA. I could never run
>>>standard kernels on mind in the 2.0 days, and distro installs would
>>>always lock up during installation, although Mandrake 8 seemed
>>>reliable so something improved.
>>I had a VIA VPX sometime ago. AFAIR it worked fine...
>>I suspect PCI conf tweaks etc could work around
>>this trouble. I'm afraid there won't be much interest
>>in fixing these oldies. For example, I got rid of that
>>board (exchanged for Socket A one) -> no way to test fixes :(
>I found a hdparm command which fixed it, though it wasn't much use
>during distro installs. It was very pleasant to see Mandrake 8 just
>work. Fwiw, Windows 95, 98 and NT4 have no problems on the box. It's
>now my "Internet Explorer 4" test rig :)
>-- Jamie

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:47    [W:0.048 / U:0.620 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site