JFFS2 mount time

Josh Boyer jdub at us.ibm.com
Fri Dec 17 12:10:31 EST 2004

Gareth Bult (Encryptec) wrote:
> Mm, my point exactly , Journaled _Flash_ File System - as opposed to
> Journaled _MTD_ _Flash_ File System .. !!
> USB flash is still flash! - just presented via a generic interface as
> opposed to an embedded one .. (AFAIK)

No, it's really not.  There is no access to the low level flash, so MTD 
cannot directly drive it.  Same principle as the Compact Flash devices, 
which MTD doesn't deal with either (other than via blkmtd).

> I guess I was sort of hoping blkmtd.c was written to enable JFFS2 to be
> more generic and embedded device independent. I suspect however it was
> written purely for testing purposes .. :(

Yes, most likely.

> As I understand it, the Flash in USB keys is identical to the MTD type
> embedded devices, except that the USB keys always come with
> micro-controllers that handle and optimise read/write/erase operations
> and present them via a generic PC/USB based disk interface. I'm not sure
> how fast MTD can be driven, but key manufacturers seem to think that
> keys should be able to run at 20Mb/sec read and 10Mb/sec write [for
> large block read/writes] which is much faster than is required for a key
> to replace a hard disk in a workstation.

The flash used in the devices is identical, but MTD has no access to it.

> I'm thinking that JFFS2 could draw in a huge additional user base if it
> catered for (or at least supported) such devices .. (!)

That's the problem.  There is no filesystem for these "abominations" 
really.  They pretend to be IDE, SCSI, etc. but are really flash 
underneath.  Neither the MTD driver or the IDE drivers (for example) 
deal with them really well, but I'm not sure that they can.

Did you try ext3 on the device with a native driver (IDE, SCSI, USB 
specific) instead of via blkmtd.c?


More information about the linux-mtd mailing list