README.falcon

苏宏峻

2023-12-01

U-Boot Falcon Mode

Introduction

This document provides an overview of how to add support for Falcon Mode to a board.

Falcon Mode is introduced to speed up the booting process, allowing to boot a Linux kernel (or whatever image) without a full blown U-Boot.

Falcon Mode relies on the SPL framework. In fact, to make booting faster, U-Boot is split into two parts: the SPL (Secondary Program Loader) and U-Boot image. In most implementations, SPL is used to start U-Boot when booting from a mass storage, such as NAND or SD-Card. SPL has now support for other media, and can generally be seen as a way to start an image performing the minimum required initialization. SPL mainly initializes the RAM controller, and then copies U-Boot image into the memory.

The Falcon Mode extends this way allowing to start the Linux kernel directly from SPL. A new command is added to U-Boot to prepare the parameters that SPLmust pass to the kernel, using ATAGS or Device Tree.

In normal mode, these parameters are generated each time before loading the kernel, passing to Linux the address in memory where the parameters can be read. With Falcon Mode, this snapshot can be saved into persistent storage and SPL is informed to load it before running the kernel.

To boot the kernel, these steps under a Falcon-aware U-Boot are required:

Boot the board into U-Boot.
After loading the desired legacy-format kernel image into memory (and DT as well, if used), use the “spl export” command to generate the kernel parameters area or the DT. U-Boot runs as when it boots the kernel, but stops before passing the control to the kernel.
Save the prepared snapshot into persistent media. The address where to save it must be configured into board configuration file (CONFIG_CMD_SPL_NAND_OFS for NAND).
Boot the board into Falcon Mode. SPL will load the kernel and copy the parameters which are saved in the persistent area to the required address. If a valid uImage is not found at the defined location, U-Boot will be booted instead.

It is required to implement a custom mechanism to select if SPL loads U-Boot or another image.

The value of a GPIO is a simple way to operate the selection, as well as reading a character from the SPL console if CONFIG_SPL_CONSOLE is set.

Falcon Mode is generally activated by setting CONFIG_SPL_OS_BOOT. This tells SPL that U-Boot is not the only available image that SPL is able to start.

Configuration

CONFIG_CMD_SPL
Enable the “spl export” command. The command “spl export” is then available in U-Boot mode.

CONFIG_SYS_SPL_ARGS_ADDR
Address in RAM where the parameters must be copied by SPL. In most cases, it is + 0x100

CONFIG_SYS_NAND_SPL_KERNEL_OFFS Offset in NAND where the kernel is stored

CONFIG_CMD_SPL_NAND_OFS Offset in NAND where the parameters area was saved.

CONFIG_CMD_SPL_WRITE_SIZE Size of the parameters area to be copied

CONFIG_SPL_OS_BOOT Activate Falcon Mode.

Function that a board must implement

void spl_board_prepare_for_linux(void) : optional
Called from SPL before starting the kernel

spl_start_uboot() : required
Returns “0” if SPL should start the kernel, “1” if U-Boot must be started.

Environment variables

A board may chose to look at the environment for decisions about falcon mode. In this case the following variables may be supported:

boot_os :
Set to yes/Yes/true/True/1 to enable booting to OS, any other value to fall back to U-Boot including
unset).

falcon_args_file :
Filename to load as the ‘args’ portion of falcon mode rather than the hard-coded value.

falcon_image_file :
Filename to load as the OS image portion of falcon mode rather than the hard-coded value.

Using spl command

spl - SPL configuration

Usage:

spl export <img=atags|fdt> [kernel_addr] [initrd_addr] [fdt_addr ]

img : “atags” or “fdt”

kernel_addr: kernel is loaded as part of the boot process, but it is not started. This is the address where a kernel image is stored.

initrd_addr : Address of initial ramdisk, can be set to “-” if fdt_addr without initrd_addr is used

fdt_addr : in case of fdt, the address of the device tree.

The spl export command does not write to a storage media. The user isresponsible to transfer the gathered information (assembled ATAGS list or prepared FDT) from temporary storage in RAM into persistant storage after each run of ‘spl export’. Unfortunately the position of temporary storage can not be predicted nor provided at commandline, it depends highly on your system setup and your provided data (ATAGS or FDT).

However at the end of an succesful ‘spl export’ run it will print the RAM address of temporary storage. Now the user have to save the generated BLOB from that printed address to the pre-defined address in persistent storage (CONFIG_CMD_SPL_NAND_OFS in case of NAND). The following example shows how to prepare the data for Falcon Mode on twister board with ATAGS BLOB.

The “spl export” command is prepared to work with ATAGS and FDT. However, using FDT is at the moment untested. The ppc port (see a3m071 example later) prepares the fdt blob with the fdt command instead.

Usage on the twister board

Using mtd names with the following (default) configuration
for mtdparts:

device nand0 <omap2-nand.0>, # parts = 9
 #: name        size        offset      mask_flags
 0: MLO                 0x00080000      0x00000000      0
 1: u-boot              0x00100000      0x00080000      0
 2: env1                0x00040000      0x00180000      0
 3: env2                0x00040000      0x001c0000      0
 4: kernel              0x00600000      0x00200000      0
 5: bootparms           0x00040000      0x00800000      0
 6: splashimg           0x00200000      0x00840000      0
 7: mini                0x02800000      0x00a40000      0
 8: rootfs              0x1cdc0000      0x03240000      0

twister => nand read 82000000 kernel

NAND read: device 0 offset 0x200000, size 0x600000
 6291456 bytes read: OK

Now the kernel is in RAM at address 0x82000000

twister => spl export atags 0x82000000
## Booting kernel from Legacy Image at 82000000 ...
   Image Name:   Linux-3.5.0-rc4-14089-gda0b7f4
   Image Type:   ARM Linux Kernel Image (uncompressed)
   Data Size:    3654808 Bytes = 3.5 MiB
   Load Address: 80008000
   Entry Point:  80008000
   Verifying Checksum ... OK
   Loading Kernel Image ... OK
OK
cmdline subcommand not supported
bdt subcommand not supported
Argument image is now in RAM at: 0x80000100

The result can be checked at address 0x80000100:

twister => md 0x80000100
80000100: 00000005 54410001 00000000 00000000    ......AT........
80000110: 00000000 00000067 54410009 746f6f72    ....g.....ATroot
80000120: 65642f3d 666e2f76 77722073 73666e20    =/dev/nfs rw nfs

The parameters generated with this step can be saved into NAND at the offset
0x800000 (value for twister for CONFIG_CMD_SPL_NAND_OFS)

nand erase.part bootparms
nand write 0x80000100 bootparms 0x4000

Now the parameters are stored into the NAND flash at the address CONFIG_CMD_SPL_NAND_OFS (=0x800000).

Next time, the board can be started into Falcon Mode moving the setting the gpio (on twister gpio 55 is used) to kernel mode.

The kernel is loaded directly by the SPL without passing through U-Boot.

Example with FDT: a3m071 board

To boot the Linux kernel from the SPL, the DT blob (fdt) needs to get prepard/patched first. U-Boot usually inserts some dynamic values into the DT binary (blob), e.g. autodetected memory size, MAC addresses, clocks speeds etc. To generate this patched DT blob, you can use the following ommand:

1. Load fdt blob to SDRAM:
=> tftp 1800000 a3m071/a3m071.dtb

2. Set bootargs as desired for Linux booting (e.g. flash_mtd):
=> run mtdargs addip2 addtty

3. Use "fdt" commands to patch the DT blob:
=> fdt addr 1800000
=> fdt boardsetup
=> fdt chosen

4. Display patched DT blob (optional):
=> fdt print

5. Save fdt to NOR flash:
=> erase fc060000 fc07ffff
=> cp.b 1800000 fc060000 10000
...

Falcon Mode was presented at the RMLL 2012. Slides are available at:

http://schedule2012.rmll.info/IMG/pdf/LSM2012_UbootFalconMode_Babic.pdf