Booting from NAND on an Armada-370 based machine -> Invalid header checksum

[Pali Rohár]

Hello!

On Sunday 05 September 2021 17:48:16 Uwe Kleine-König wrote:
> Hello,
> 
> I'm trying to unbrick a Netgear ReadyNAS 104 (Armada 370). (I
> accidentally erased the u-boot partition in NAND when I tried to change
> the NAND partitioning to make the Debian bullseye kernel+initramfs fit.)
> 
> I have the Vendor U-Boot image that I can boot using kwboot. Its first
> byte is 0x8b which is the right for NAND booting. The image's size is
> 0xda548 bytes.
> 
> I wrote the image to the start of NAND and verified it to be correctly
> written:
> 
> 	Marvell>> nand read 0x2100000 0 0xda548
> 
> 	NAND read: device 0 offset 0x0, size 0xda548
> 	 894280 bytes read: OK
> 
> 	Marvell>> dhcp
> 	egiga1 wait for link .Done - link up.
> 	...
> 	DHCP client bound to address 192.168.77.145
> 
> 	Marvell>> tftp 0x2000000 192.168.77.175:u-boot-rn104-2.0.img
> 	Using egiga1 device
> 	TFTP from server 192.168.77.175; our IP address is 192.168.77.145
> 	Filename 'u-boot-rn104-2.0.img'.
> 	Load address: 0x2000000
> 	Loading: #############################################################
> 	done
> 	Bytes transferred = 894280 (da548 hex)
> 
> 	Marvell>> cmp.b 0x2000000 0x2100000 0xda548
> 	Total of 894280 bytes were the same
> 
> There are no bad blocks in this area.
> 
> Also the checksum is right as far as I understand:
> 
> 	$ python3
> 	...
> 	>>> a = open("/srv/tftp/u-boot-rn104-2.0.img", "rb")
> 	>>> data = a.read(0x14000)
> 	>>> len(data)
> 	81920
> 	>>> hex(sum(data[:31]) + sum(data[32:]))
> 	'0x79f616'
> 
> So the checksum field should be 0x16 (at offset 31), which it is:
> 
> 	Marvell>> md.b 0x2100000
> 	02100000: 8b 00 00 00 48 65 0c 00 01 01 00 40 00 40 01 00    ....He..... at .@..

So...
Image version = 0x01
Header Size MSB = 0x01
Header Size LSB = 0x00 0x40 = 0x4000

So header size is 0x014000 = 81920

> 	02100010: 00 00 00 00 00 00 00 00 00 02 01 00 00 00 01 16    ................

Checksum = 0x16

Which seems that the header checksum is correct.

> 	02100020: 02 01 18 35 02 00 00 00 5b 00 00 00 00 00 00 00    ...5....[.......
> 	02100030: ff 5f 2d e9 1c 00 00 fa 00 00 a0 e3 ff 9f bd e8    ._-.............
> 
> Still when trying to boot I get:
> 
> 	BootROM 1.08
> 	Booting from NAND flash
> 	BootROM: Invalid header checksum
> 	BootROM: Bad header at offset 00010000
> 	BootROM: Bad header at offset 00020000
> 	BootROM: Bad header at offset 00030000
> 	BootROM: Bad header at offset 00040000
> 	BootROM: Bad header at offset 00050000
> 	BootROM: Bad header at offset 00060000
> 	BootROM: Bad header at offset 00070000
> 	BootROM: Bad header at offset 00080000
> 	BootROM: Bad header at offset 00090000
> 	BootROM: Bad header at offset 000A0000
> 	BootROM: Bad header at offset 000B0000
> 	BootROM: Bad header at offset 000C0000
> 	BootROM: Bad header at offset 000D0000
> 	BootROM: Bad header at offset 000E0000
> 	BootROM: Bad header at offset 000F0000
> 	BootROM: Bad header at offset 00100000
> 	BootROM: Bad header at offset 00110000
> 	BootROM: Bad header at offset 00120000
> 	BootROM: Bad header at offset 00130000
> 	BootROM: Bad header at offset 00140000
> 	BootROM: Bad header at offset 00150000
> 	BootROM: Bad header at offset 00160000
> 	BootROM: Bad header at offset 00170000
> 	BootROM: Bad header at offset 00180000
> 	BootROM: Bad header at offset 00190000
> 	BootROM: Bad header at offset 001A0000
> 	BootROM: Bad header at offset 001B0000
> 	BootROM: Bad header at offset 001C0000
> 	BootROM: Bad header at offset 001D0000
> 	BootROM: Bad header at offset 001E0000
> 	BootROM: Bad header at offset 001F0000
> 	BootROM: Bad header at offset 00200000
> 	BootROM: Bad header at offset 00210000
> 	BootROM: Bad header at offset 00220000
> 	...
> 
> Is there anything obvious I'm missing or doing wrong? Does "Invalid
> header checksum" indicate that the 0x16 is wrong or is there another
> checksum anywhere that I missed? Any other idea?

Image version 1 has only one header checksum (at 0x1F) and one data
checksum (4 bytes at the end of data image). But data checksum can be
verified only after loading data image to the RAM, which can happen only
after executing binary headers. If you are using standard DDR training
(prevent in binary headers) then it should print some debug log on UART.
And if you do not see it then error should refer to header checksum at
0x1F offset.

Could you run latest version of 'mkimage -l' from U-Boot git master on
your image to verify that image is really valid?

> If you have an idea and prefer irc over email, I hang around e.g. in
> #mvlinux on libera as ukleinek.
> 
> Best regards
> Uwe