Hi all,

The last week I finished to build my AT91SAM9260 based custom board. I have some problems and questions, I hope someone will have idea because I am blocked since 2-3 days.
My board is largely inspired from the AT91SAM9260-EK board.
The 3 majors differences are :
1) My 64MB SDRAM are two MT48LC32M8 chips, connected with 16 bits databus to the processor. On my board I need a lot of GPIO thus I need D16 to D31 to be GPIO
2) My Nand flash is a 512MB (4Gb) from ST (NAND04GW3B2DN6E). The ready/busy pin of the Nand is connected to PB16 and not PC13.
3) My ethernet Phy is a DP83848I from National, and the interrupt line is connected on PA22 (this phy has no ETXER, I don't know why).

On my board I want to reserve the Nand memory to store kernel image and rootfs, thus the bootstrap, uboot and uboot's environnement are stored in the dataflash, a AT45DB161 connected on NPCS1.

Here is what I done in the bootstrap sources :

- Modify the sdram init procedure to use a 16 bits bus and do not set D16-D32 as peripheral mode
- Change debug baudrate to 57600 bauds (I don't know why but my pcmcia/seriall adapter on my laptop don't work at 115200)

My modification works well I am able to load u-boot in SDRAM and execute him successfully, thus I have not problem with the bootstrap

Here is what I done in the uboot sources :

- Modify board/at91sam9260ek/nand.c to use PB16 and not PC13 as ready/busy input pin
- Modify include/configs/at91sam9260ek.h to say uboot environnement is in dataflash
- Modify AT91C_SPI_CS1_CLK in cpu/arm926ejs/at91sam926x/spi.c to have 5Mhz instead of 33Mhz, because my AT45DB161 was not found (This dataflash works at max 13Mhz), and change baudrate to 57600 bauds (I don't know why but my pcmcia/seriall adapter on my laptop don't work at 115200)
- Create a DP83848.c and DP83848.h to support my Phy. In fact I copied the dm9161a.c and dm9161a.h and I just modified names and the OUI value in the DP83848.h

My modications works well, nand flash is detected, dataflash is detected, and Phy is detected.
I can save uboot env successfully in dataflash. I can also successfully download kernel and image by tftp, but I found that the download speed is not very rapid. I download a 13MB rootfs image in 12-13sec, thus approx 1MB/s. Is it a normal download speed ? Someone can tell me the dowload speed of his board ?

Here is what I done in the linux kernel sources :

- First I installed official 2.6.25 kernel sources, and applied 2.6.25-at91.patch.gz patchs from linux4sam.org
- I done a make mrproper, and a make ARCH=arm at91sam9260ek_defconfig
- I compiled the kernel with a make ARCH=arm CROSS_COMPILE=arm-linux- , to see if everything was OK. The compilation was successfull.
- After I modified the kernel configuration (make ARCH=arm xconfig) to activate MTD and nand flash support with Hardware ECC, MTD partionning support, ext2 support, ext3 support, and jffs2 support.
- I changed the boot cmd line to : mem=64M console=ttyS0,57600 root=/dev/mtd1 rw
- I modified the arch/arm/mach-at91/board-sam9260ek.c file to change the nand ready/busy pin(PB16 in my case) and the PHY interrupt pin (PA22 in my case).

For now I don't put any root filesystem in the flash, the first thing I want to do is to verify that kernel boot correctly and see my nand flash.

Here is what I do :

U-Boot> nand erase

--> Reboot

U-Boot> nand bad
Device 0 bad blocks:
0d820000
0d860000

U-Boot> tftp 21000000 uImage
--> it's my compiled kernel, I done a mkimage on it.

U-Boot> bootm 21000000
--> kernel startup :



My nand flash is detected correctly. But I have a problem, the nand scan for bad block listing is very long, it takes more than 60sec !! uboot only takes 3-4sec to do that. Why scanning is very long and how reduce this time ?

Now I want to put a jffs2 file system in the nand flash but I have a big problem, every time I write something in the flash from uboot, the blocks are seen as bads a the next uboot restart or even in the kernel message when nand is scanned !!

Example :



-->reboot

U-Boot> nand bad



And if for exemple I do a nand erase clean, at the next uboot restart all blocks are marked as bad if I launch the "nand bad" command, or if I boot the linux kernel.

Any idea where the problem can come from ?
Sorry for my long message and my poor english

To resume my question :
- What is your dowload speed with tftp in uboot ?
- Why the bad block scanning is very long in linux (>60sec) compared to uboot (only 3-4sec)
- Why blocks are considered bad at next startup when I write into it with uboot (and ther is no write error !)

Regards

I just tried to load my rootfs.jffs2 image with samba. After that, u-boot doesn't see that blocks are bads.
Thus I have a problem with the uboot write function. It tells the write in nand flash is OK but it mark block as bad, how it is possible !!??

Because with samba I have a valid jffs2 filesystem in nand flash at offset 0x40000 (mtd1), I try to boot linux :



Anyone knows why the kernel is unable to mount the mtd1 partition ?
Regards

Hi nlc.
* We also make a custom board for the AT91SAM9260 but with minor differences respect the EK. The version of u-boot that we are using is 1.3.0 with support for the DP83848 (please see mi post in the thread "U-Boot ethernet phy DP83848 support").
* For the question about the tftp speed, in my board I get the same speed as you: 13Mb in 13/15 seconds.
* At first we have problems with the nand flash. u-boot and linux kernel don't read the nand ID right, so I have to edit the code of u-boot in file nand_base.c, function nand_scan, and add a reset command before the read_id command. This solve our problem with the nand.
* Using the angstrom rootfs the nand are working well. With the buildroot version I had problems with blocks size of the jffs2 image, etc, so for now I'm using a usb key to mount the filesystem. If you can do that may be a good way to make progress until the nand still with problems.
* Also you can put the rootfs in a mmc/sd card or mount the fs via nfs, but I don't try these options yet.

I hope this may help you. Sorry for my poor english too!

For the angstrom demo the kernel command line arguments are:
mem=64M console=ttyS0,115200 root=/dev/mtdblock1 rw rootfstype=jffs2

Note the use of mtdblock1 instead mtd1. Try with this change maybe help.

Thank you for the download rate mesurement
I will pass on uboot 1.3.0 when all my problem will be solved, because for now it's OK with DP83848.

For the nand ID or nand detection, uboot/linux has always detected it correctly.

With samba I loaded the angstrom filesystem in the nand (because uboot nand write problems...), but I my linux kernel doesn't want to see it

The kernel correctly detect the nand and partitions :



But it's as is the kernel doesn't have created the /dev/mtdblockX devices, because they are not listed as available partition :

Yes I tested too (mtdblock1 / rootfstype option), same result

And what is your nand size and how many time takes your bad block scanning in kernel startup ?

The nand in my board is MT29F8G08MAAWC, 8gbits (1Gbyte), 8 bits data width.
My jffs2 image generated with buildroot don't work, but I don't try changing the jffs2 parameters (block size, etc) yet. With the linux4sam demo all is ok. The boot completes in one minute, but the bad blocks scanning its not so long.

I will try to find the explanations because I don't understand why linux scanning takes 60sec to scan 4096 block and u-boot takes 3-4 sec !!

gastu, I tried the uboot 1.3.0 you have used too (I used your link).

it's OK for DP83848, but another kind of problem appear with my nand flash
Now my nand isn't detected correctly :



Chip IDs are correctly read (0x20 and 0xDC), but not correctly associated with the manufacturer "ST Micro" as before. And pagesize / blocksize / OOBsize are wrong (must be 2048 / 128K / 64) !

I try to locate the problem....

Ok I have u-boot working with nand flash.

In drivers/nand/nand_ids.c there is 2 Nand device with the same device ID :

{"NAND 512MiB 3,3V 8-bit", 0xDC, 512, 512, 0x4000, 0},
{"NAND 512MiB 3,3V 8-bit", 0xDC, 0, 512, 0, NAND_SAMSUNG_LP_OPTIONS | AND_NO_AUTOINCR},

When uboot scan the table it stop on the first....where page size and block size are hard coded in the table.
I commented the first one, and now uboot detect correctly the nand because page size is not hardcoded in the table and uboot reads another register in the nand to get the informations :



Now write to the Nand from uboot works, after a reboot blocks where uboot has wrote something are not considered bads

I don't know why the manufacturer Name doesn't appear, bur for now it's not really important


Now I need to undestrand why linux doesn't create my mtdblocks device Surelly a kernel option but I don't see that option

OK I found why bad block scanning was very long during kernel start.

In fact I done a mistake when I modified my rdy_pin in the arch/arm/mach-at91/board-sam9260ek.c file

Thus the ready pin was always at low level and thus in this case the the function nand_wait_ready() in driver/mtd/nand_base.c in limited to 2 "jiffies", thus 20ms.....

Now I configured correctly my rdy_pin and the bad block scanning takes (looked on the scope)......200ms !!!

Next step have kernel able to create mtdblock devices....

I progress !
Now it's OK mtdblock devices are created and my mtdblock1 is successfully mounted. It contains a Angstrom jffs2 root filesystem. The secret was to activate the "Caching block device access to MTD device" option....

But problem, I have no console nor messages :



After that, that all, no console ?
Curiously the board is not died because I have echo characters on minicom !!

Hi gasty,

Now I have exactly the same problem. I successfully finished an entire buildroot process, but the jffs2 image doesn't work. I tryed with my kernel 2.6.25 built alone or with the buildroot kernel (2.6.24-4), and result is the same, the kernel can't mount the filesystem :

Hi nlc.
I think the problem of magic bitmask 0x1985 are related to the parameters that buildroot passes to the mkimage program to generate the jffs2 image. These parameters are given by default for the nand in the at91sam9260ek. See in the .config file for buildroot:

BR2_TARGET_ROOTFS_JFFS2_PAGESIZE=0x420 (1024+32 bytes)
BR2_TARGET_ROOTFS_JFFS2_EBSIZE=0x2100 (128K + 4K bytes=0x21000. One '0' less?)

The values for my nand are:
BR2_TARGET_ROOTFS_JFFS2_PAGESIZE=0x840 (2048+64 bytes)
BR2_TARGET_ROOTFS_JFFS2_EBSIZE=0x4400 (256K+8K bytes=0x44000)

See your nand specs.