4K blocks of zeros in JFFS2 filesystem

[Jeremy Slater]

I am having file corruption problems on my JFFS2 partition where
eventually 4K blocks of files will end up as 4K blocks of zeros.  I am
running a 2.6.12 kernel on an at91rm9200 processor with 128MB of NOR
flash spanning two concatenated chips.

I am running a simple test to recreate this issue: 10000 static files
are periodically scanned via md5sum.  Another process fills the file
system with data until there is only 2MB free, deletes all the files
and repeats.  After 1 to 10 hours there will be a small number of
files failing the md5sum and containing the blocks of zeros.

I am still working to collect more kernel output, but here is a
snippet that appeared up a short time before a corrupt file was
detected:

Node CRC 4c4e2482 != calculated CRC b69a14b7 for node at 0214000c
Node CRC 2e5eb6c0 != calculated CRC f4ebfeae for node at 01e4000c
Node CRC ae8a5503 != calculated CRC 59093d5e for node at 000bdfa0
Node CRC 4081f13e != calculated CRC 216eb339 for node at 0168000c
Node header CRC failed at 014a000c. But it must have been OK earlier.
Node was: { ffff, ffff, ffffffff, ffffffff }
Node CRC 10e840d9 != calculated CRC 5ee4806a for node at 0147e3b0

You will notice that I also get a lot of CRC errors, which I am not
sure if that is normal.

This test is reliably failing with a 112MB mtd partition.  I have run
variations of this test on smaller (12MB to 64MB) mtd partitions and
have not seen any failure yet.  I have also run the test on a 100MB
mtdram partition and have not seen any failures.

Does anyone have suggestions on what I can do to uncover the source of
this corruption?

Thanks,
Jeremy