mounting "unformatted" nandflash

[Ricard Wanderlof]

Hi all,

Before I start digging into this myself, perhaps someone here has an 
answer.

We are running an embedded system with 128 megs of nand flash 
(large-block). To put it simply, our scenario is that we mount a large 
portion of empty flash as a jffs2 partition. Since the partition is empty, 
there are no [oob] cleanmarkers, and upon mounting, jffs2 scans the 
partition, marks all blocks as needing erasure; this is then ultimately 
controlled by the pdflush kernel task.

The strange thing is that pdflush then takes almost 100% cpu for a minute 
or so to complete. Erasing the whole flash (e.g. using flash_eraseall) 
just takes a few seconds, writing the cleanmarkers should take even less. 
So why is pdflush taking so much time to complete it's task? I've figured 
that it must be the erase/cleanmarker-write process which is the culprit, 
as upon rebooting pdflush exhibits no such behavior when the cleanmarkers 
are in place).

(Yes, in this case we could use flash_eraseall -j (which incidently 
doesn't seem to work but could be easily fixed) to 'format' the partition 
in question. However, for various reasons this isn't practical for us as 
the initial flash write is done using a non-linux tool).

/Ricard
--
Ricard Wolf Wanderlöf                           ricardw(at)axis.com
Axis Communications AB, Lund, Sweden            www.axis.com
Phone +46 46 272 2016                           Fax +46 46 13 61 30