IDE data corruption.

From: joe briggs (
Date: Tue Jun 17 2003 - 11:06:34 EST

I have sent several postings over the last few weeks with data supporting a
potential weakness in the IDE driver that results in file system corruption
on systems with heavy PCI activity (specifically, video capture for
surveillance systems). Affected hardware includes Tyan 2460/6 dual athlon
motherboards, GigaByte GA7VXP athlon, and Aopen G4MX P4/2.4 motherboards.

On all of these systems running Debian 2.4.19 and 2.4.21 kernels, data
corruption would occur when sustaining an aggregrate video capture and
transfer rate of approximately 40 320x240 YUV (12 bit/pixel) frames per
second. Test systems where not runing X, and used ReiserFS. All systems had
one IDE boot/system drive, and a 2-drive IDE raid using Promise or 3-ware pci
controllers. The frame-grabbers used the bttv driver. The RAID was used for
the frame data and database.

Eventually, the system will just freeze while trying to swap. Errors present
on the console or system log, if any, where "memory.c: 100: pmd", a plethora
of ReiserFS errors, and/or a page fault. Occassionally we would see a 'hda
missed interrupt" in there. Reboot and subsequent remount of the file
systems frequently failed, and would require complete rebuilding as
reisferfsck --fix-fixable or -rebuild-tree would not clean it all up.

Shiftng the RAID from Promise to 3ware stopped corrption on the data drives,
limiting it to the system drive.

I then modified the fstab on the system drive to mount the 3ware as the root
filesystem, eliminating all data access to the /dev/hda drive after startup,
and it eliminated the file system corruption completely.

The 3ware controller is a scsi controller that reads and writes to ide drives.
Whereas the Promise is just anlother ide chip (my understanding).

I tried these experiments under both Debian/2.4.19/ReiserFS and RedHat
7.3/ext3 with similar results, so I don't think that it is a filesystem

It is my opinion based on this data that EITHER the IDE hardware misses
interrupts or can corrupt data under heavy PCI DMA activity, OR the driver is
not detecting or handling issues correctly.

I am available to test any patches or ideas that the maintainers might be so
kind as to offer.

Joe Briggs
Briggs Media Systems
105 Burnsen Ave.
Manchester NH 01304 USA
TEL/FAX 603-232-3115 MOBILE 603-493-2386
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

This archive was generated by hypermail 2b29 : Mon Jun 23 2003 - 22:00:21 EST