[lkml]   [2003]   [Jun]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectFS Corruption with VIA MVP3 + UDMA/DMA
I was wondering if anyone knows about this issue or has had this problem?

I've been fighting FS corruption since switching to a UDMA hard drive
(Maxtor) on my FIC PA2013 with the VIA MVP3, but I didn't really know
that since the change was a result of a dying drive. Finally, through
the chance of having installed an older Quantum drive that only allowed
DMA MultiMode 2 as the fastest mode, I found the problem.

At least on my board, the IDE UDMA/DMA implementation appears flawed
[lspci gives: "VT82C586/B/686A/B PI (rev 6)"]. I've had the same
problem with three UDMA DVD drives -- two Toshibas and a Pioneer. They
would all lock-up installing software or corrupt the data being
installed causing the installation to fail. I've also had four other
Maxtor hard drives (3 factory certified, one retail) randomly corrupt
content on the drives (fs type doesn't matter [NTFS, FAT, FAT32, EXT2,
REISERFS have been tried]). That means that every UDMA drive I've
plugged in has had data corruption issues (trying no less than 10 IDE
cables, which I confirmed good on the Promise and I've tried both IDE
channels on the MVP3). I started with Linux kernel version 2.2 and the
problem remains up to 2.5.73. I've also confirmed the issue in Windows
98SE, Windows XP and Windows 2003 RC2. At this point, one might as well
stick in a PCI adapter, since with "hdparm -t" I get between 5.5-8.5
MB/s. With the same mode on my Promise 20269 I get 12 MB/s , so clearly
something is odd. I know the hard drive can do better since I can get
20MB/s in UDMA 2 on the MVP3, but of course that's not safe.

Even at DMA multiword 2, I can force r/w errors by heavy io. Moving to
DMA mode 1 clears up the errors, but performance degrades to a
consistent 5.49 MB/s (all the higher modes actually vary 2-6 MB/s
between runs of hdparm) while the Promise in the same mode still gets 12
MB/s consistently. I've found that copying a 300 MB file to my drive on
the Promise, making 12 new files while making 12 duplicates to the drive
on the MVP3 can still force errors in dma multiword 2. I check for
write errors by comparing the copied files to the source file. It's
much easier to create errors in any of the UDMA modes. Errors actually
seem more likely to occur during actual use of the system since they are
fairly common even in multiword 2, but the copying method makes it
extremely repeatable (though not all the files are corrupted, that
part's random).

My current configuration has the Promise as the boot device with a
single drive on the primary. I have my DVD (UDMA) on the secondary of
the Promise. My other Maxtor hard drive is on the primary channel
(alone) of the MVP3 with UDMA disabled in the bios (and thus not used by
WinXP/2003 or Linux). Finally my cd writer and IDE Zip are on the
secondary channel of the MVP3. I'd put my hard drive on the secondary
channel of the Promise, but for some reason the computer won't boot with
both hard drives on the Promise (even though they're on different
cables)... I don't remember right now if it just locks up during OS
loading or if it won't post. I can test it if that information is required.

Otherwise my machine is fairly stable (3+weeks, but I usually have to
boot to Windows for Illustrator and such before then) and I don't get
any corruption when copying files around on my Promise controller. I
can get errors even copying from my MVP3 drive to itself (making
defragging dangerous on my shared FAT32 partition) even in DMA multiword 2.

Is there anything that can be done with the IDE drivers for this chipset
to make it "safe" without resorting to forcing DMA mode 1?

Wes Janzen

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 13:46    [W:0.105 / U:14.688 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site