Re: Repeatable File Corruption (ECS K7S5A w/SIS735)

[Andre Hedrick <andre@aslab.com>]

Greetings dan,

Well no kidding !!!
Just maybe it is the VM layer getting it wrong again.
Please try an -ac kernel and see if it still exist.

As most know that there are corrections there that have not been accepted
into the main tree ..... OUCH that hurts ....

Cheers,

On Thu, 25 Oct 2001, dan wrote:

> 
> ===== SUMMARY
> 
> I have found a file corruption issue when writing files seemingly related
> to my new motherboard.
> 
> 
> ===== HARDWARE
> 
>   It is repeatable and verified on other boards of the same model.  This
> just started happening when I upgraded the system.  The following is a
> link to the ECS K7S5A board in question, the SIS735 chipset, and a
> hardware description:
> 
>   http://www.ecsusa.com/ecsusa/www.ecs.com.tw/products/k7s5a.htm
>   http://www.sis.com/products/chipsets/oa/socketa/735.htm
> 
>   ECS K7S5A motherboard
>   AMD 1.4Ghz Tbird
>   2x256Mb Micron DDR PC2100
>   hda: ST36421A, ATA DISK drive
>   hdb: QUANTUM FIREBALLP LM10.2, ATA DISK drive
>   hdc: IC35L060AVER07-0, ATA DISK drive
> 
> 
> ===== DESCRIPTION
> 
> The bug was first noticed when running a game server that verifies client
> map files with the server map files.  Some of the maps, no matter what I
> did, were always different, causing the server to reject clients.  Using
> the commands md5sum and cmp, I noticed that each time I copied a new file
> into place, it was slightly different:
> 
>   $ cp de_aztec.bsp de_aztec.bsp2
>   $ cmp de_aztec.bsp de_aztec.bsp2
>   de_aztec.bsp de_aztec.bsp2 differ: char 2287174, line 6590
>   $ md5sum de_aztec.bsp de_aztec.bsp2
>   c138a969edf84abcfc5fe4b4b2a2c5ed  de_aztec.bsp
>   36b5bdd244d69a8d0fb3c81ba25e1df1  de_aztec.bsp2
> 
> Not only that, every copy is different from every other copy:
> 
>   $ cp de_aztec.bsp de_aztec.bsp3
>   $ cp de_aztec.bsp de_aztec.bsp4
>   $ cp de_aztec.bsp de_aztec.bsp5
>   $ md5sum de_aztec.bsp de_aztec.bsp?
>   c138a969edf84abcfc5fe4b4b2a2c5ed  de_aztec.bsp
>   fe858630be18e37eba8ea73a6be941c9  de_aztec.bsp2
>   87bd1ad8b0ff2c58ebff5372f865fa35  de_aztec.bsp3
>   7cffbffca0600c80e6cae336f52c885b  de_aztec.bsp4
>   91287ef7c8eb047d8f305315b4e0ce55  de_aztec.bsp5
> 
> It does not seem to be an issue with reading from the drive because the
> md5sum of the original file is always the same.
> 
> This is repeatable with any command that will write a file out including
> tar, rsync, and ftp.  If I move the file within the partition, there is no
> corruption.  Copying the file one character at a time works.
> 
> This does not happen with every file on the drive.  In fact, out of the
> 1400 files related to the server, I only found a few dozen that I could
> repeat this issue with.  I also have complaints from other users of the
> system of files being corrupt and have found a number myself.
> 
> I have repeated this with the following:
> 
>   Duplicate systems (I have 5)
>   Multiple hard drives (IBM, Quantum, Seagate)
>   Different filesystems on multiple drives (reiserfs, ext2)
>   With and without different drivers in the kernel (IDE chipset support)
>   Kernels 2.4.10 through 2.4.13
>   Drives running at different speeds  (ATA-100, 66, 33)
>   Slow processor (AMD 1.2GHz Tbird)
> 
> 
> ===== WORKAROUND
> 
> The problem only went away when I replaced the motherboard.  I also
> haven't had any file corruption issues running Windows2000 on the same
> hardware with the same files.  I moved all of the hardware in the original
> system to a new motherboard (ASUS A7A266)  and the problem went away.
> 
> I have CC'd the IDE chipset maintainer because I can only assume it might
> be related.
> 
> Thank you for fighting through this email, I know it was long and boring.
> 
> 
> dan
> 

Andre Hedrick
CTO ASL, Inc.
Linux ATA Development
-----------------------------------------------------------------------------
ASL, Inc.                                     Tel: (510) 857-0055 x103
38875 Cherry Street                           Fax: (510) 857-0010
Newark, CA 94560                              Web: www.aslab.com