Jatin Sharma wrote:

I am running U-Boot version U-Boot 1.3.1-rc1 on my board that has a Broadcom network switch. My goal is to initialize this switch at the bootloader level and be able to tftpboot a Linux kernel from one of the ethernet ports of this switch.

As _I_ understand, there are two ways to initialize the switch from u-boot.

First

Theory: Link the switch driver software with the u-boot.bin and flash it on the board. Switch driver's entry point would be called somewhere in u-boot to initialize it. Size of the u-boot.bin will increase and the u-boot partition on the NOR flash will need to be updated to accommodate the new binary size.

Practice: After I linked the driver software with the u-boot.bin, the board didn't boot any further after u-boot got relocated to RAM with this following message: "Now running in RAM - U-Boot at: 0ff6d000"

I did the following to link switch driver library (MDK_BLIBS) with u-boot in Makefile:

$(obj)u-boot: ......... --start-group $(__LIBS) $(MDK_BLIBS)--end-group $(PLATFORM_LIBS) \ .........

Switch library is compiled with same TEXT_BASE and other flags as that of u-boot.

Second

Theory: Build the network switch driver as a u-boot application, flash it on a free partition on the NOR and after u-boot has booted, it will somehow do the following to initialize the network switch.

cp.b 0xfff90000 0x40000 0x5bfac go 0x00040418

Practice: The above two commands worked to initialize the switch. I still need to understand how to run these command automatically after the u-boot has booted.

Questions: 1.) Which design is recommended, first or the second? Second is easier for me to implement as I don't have to change the partition sizes but could easily be overlooking some obvious problems.

It would help to know which type of Broadcom switch it is you're using. A lot of them have a SPI-based control plane, in which case you can set up a simple port configuration without an elaborate driver.

regards, Ben