JFFS2 Flash Filesystem corrupted

JFFS2 Flash Filesystem corrupted

[Billalabeitia, Jose Carlos (GE Consumer & Industrial)]

	Hi:
	
	I have a custom board with the following features:

		CPU MPC860T (PowerPC)
		16 MB of FLASH (2 29LV650UE (4Mx16 bit) compatible JEDEC & CFI )
		16 MB of RAM
		linux-2.4.4-2002-02-14-denx
		linux-2.4.4-jffs2-2002-05-01.patch for the JFFS2 (basically improve GC performace)

	We have experienced in a very few occasions, in fact only twice or three times, a weird problem.
Our application are composed of multiple excutable and module files that are started and insmoded 
respectively one by one during the boot up process. After a system rebooting everything was fine
and we got the app running, therefore executable and module files were right. 
However one day after a rebooting we realized that the app. didn´t go and we check the integrity of the
excutable and module files by comparing them with the original ones, and we discovered that 4Kbytes of 
executable module file were replaced with all zeros. 

	Due to the low probability of this type of problem that we have encountered it makes me think
that it might be a hardware problem, but in this case I can´t find a good explanation for it.

	Another thing that surprises me is that the filesystem says that that file is OK, so I suppose
then that the CRCs of the nodes were the zeros are located are correct, so my suspect is that those
zeros were put intentionally.

	On the other hand that file never changes so I think that the only one responsible for moving
those data would be the GC when the "wear levelling" process. Any ideas?

									Thank you
										Billa

Re: JFFS2 Flash Filesystem corrupted

[Thomas Gleixner]

On Tue, 2004-07-20 at 20:18, Billalabeitia, Jose Carlos (GE Consumer &
Industrial) wrote:

> linux-2.4.4-2002-02-14-denx
> linux-2.4.4-jffs2-2002-05-01.patch for the JFFS2 
> (basically improve GC performace)

Sorry, I don't have any clue which state of mtd/jffs2 is incorporated in
those kernel versions. Deducing from the date 2002-xxx I assume both the
kernel and the mtd/jffs2 code are so _ANCIENT_ that it will be hard to
figure out if there are bugs which are solved in meantime or give you
any meaningful advice how to track down those problems.

tglx

Re: JFFS2 Flash Filesystem corrupted

[David Woodhouse]

On Tue, 2004-07-20 at 22:08 +0200, Thomas Gleixner wrote:
> On Tue, 2004-07-20 at 20:18, Billalabeitia, Jose Carlos (GE Consumer &
> Industrial) wrote:
> 
> > linux-2.4.4-2002-02-14-denx
> > linux-2.4.4-jffs2-2002-05-01.patch for the JFFS2 
> > (basically improve GC performace)
> 
> Sorry, I don't have any clue which state of mtd/jffs2 is incorporated in
> those kernel versions. Deducing from the date 2002-xxx I assume both the
> kernel and the mtd/jffs2 code are so _ANCIENT_ that it will be hard to
> figure out if there are bugs which are solved in meantime or give you
> any meaningful advice how to track down those problems.

Indeed. Please try to reproduce in 2.6.8-rc2

-- 
dwmw2

RE: JFFS2 Flash Filesystem corrupted

[Thomas Gleixner]

On Tue, 2004-07-27 at 12:45, Billalabeitia, Jose Carlos (GE Consumer &
Industrial) wrote:

Can you please use a mailer which does appropriate formatting !

> I made a test with the JFFS2:
> - Compile the JFFS2 in debug mode level 1.
> - All the messages sent to the console were captured in a file.
> - Run an app. test that sent SIGUP signal to the second partition
>   of the JFFS2 (where the code resides) every 5 seconds.
> - Left it running overnight (14 hours approx.)
>
> What I got was after rebooting the system this morning:
> 
> No anormal message appeared during the mounting process 
> of JFFS2 second partition.
> One module file had changed; 400 bytes (not 4Kb) were replaced 
> with zeros.
> The time and the size of the modified module file were unchanged. 
> So I suspect of the GC when moving nodes or the buffer cache 
> (as you said).
> 
> All of this make me reinforce the idea that the problem has to 
> do with the GC
> with the "wear levelling", because I think that it is the only 
> one that can move nodes 
> of a file that never changes without modifying its time and size, 
> and with the correct CRC.

That's pure speculation. 

Check the captured debug messages. They contain the answer to your
question. It make take some time to figure it out, but you can
definitely find the place where the file is touched / modified / moved
whatever.

tglx

JFFS2 Flash Filesystem corrupted

["畠中 啓 / Hajime HATANAKA"]

On Tue, 2004-07-20 at 20:18, Billalabeitia, Jose Carlos (GE Consumer &
Industrial) wrote:
> linux-2.4.4-2002-02-14-denx
> linux-2.4.4-jffs2-2002-05-01.patch for the JFFS2 
> (basically improve GC performace)

We have similar problem on
linux-2.4.22-jffs2-2004-8-xx with different CPU
and NAND flash memory.

Do we need try with latest kernel?

Have this problem been solved finally?