Hi,
I work on an AT91RM9200-DK based product.
I was used to boot from a filesystem on a nfs server but, now, I want to boot from my SPI DataFlash.
In ext2 format, my filesystem's size is arround 13M but I have only 5M of free SPI DataFlash space. When I compress it in ext2.gz format, my filesystem's size decrease to 3M. It's just fine but I have some questions :
- can I boot from a ext2.gz filesystem which is store on SPI DataFlash... ?
- if yes : how ? What are the arguments needed by the kernel ?
- is there a more adapted format ?
Thanks by advance,
Joran

Hi Joran,

do not use ext2 format. SquashFS, cramfs or jffs are your friends. Together with a build in MTD device driver you can boot this filssystems with root=/dev/mtdxxxx kernel parameter.

best regards
Jens

Hi buser,
Thanks for this answer, I haven't seen it before...
But I don't know how I can create the MTD device in the /dev directory for my SPI Dataflash...

There are different ways to boot linux. You can either load a init ramdisk from bootloader to SDRAM and boot with that (and the initrd has then enough utilities to hook to some bigger storage and continue) or let the kernel mount the root from the device directly.
Secondly. To get mtd devices, you must specify them on kernel command line. Divide your dataflash to kernel and filesystem (maybe multiple partitions are needed), calcluate the offsets and lenghts of each (align them to erase block size) and construct the kernel command line parameter mtdparts= accordingly. For example, my flash layout is usually [bootloader, bootloader config, kernel, ro filesystem, rw filesystem] and in linux it is mapped as [whole flash, bootloader, bootloader config, kernel, ro filesystem root, rw filesystem area] Set the root= kernel command line parameter to the right partition, the first is mtd0, second mtd1 and so on (in my example, 4). If you boot the kernel without initfs, you must compile in the kernel the drivers needed for that filesystem (not as modules).

Now, choose the filesystem that is the best for you. Either use read-only (but better compression) squasfs, read-write jffs2 etc. Usually you need to mount /tmp as tmpfs in your init scripts, sometimes /var and /etc, too (or just symlink what's needed to be rw to /tmp/something).

Last, you must edit your scripts to create the filesystems. There are tools, that create your FS image. Some tools need root privileges to be able to create an image from the exact tree on your host computer; some tools allow the use of some configuration files to create device nodes and set required permissions while creating the image just on the image (so you don't need root access on host pc to set them earlier)

A big thanks for this helpful response.
I have changed my kernel and everything is working now.
I can boot with read only cramfs or jffs2.
Thanks again for this response, I think that it will be useful for many others peoples !

Hi Joran,

Can you provide some details on the steps you got the jffs2 system onto the NAND Flash and how you got that filesystem mounted ?

For example, upon kernel startup, how to write the jffs2 file into the NAND flash.

Thanks in advance.

Second that, and additionally;
compared to block devices as regular harddrives(like the ones I am used to), you need to write a partitioning scheme to be able to write the filesystems and later mount them.
How are the created?

I think you need to edit the board support source in the kernel to create partitions. E.g. http://fxr.watson.org/fxr/source/arch/a ... x-2.6#L130 for the 9260ek. Then you just use mknod like you would for any other device in /dev/. The char devs have a weird minor number scheme, they alternate rw and ro, with each physical partition getting 2 minors.



The above is from a non-at91 system, but the mtd layer works the same. You can see how the major and minor numbers relate to the partitions.

Interesting. Thank you