Hi,
On 02/07/2015 04:17 AM, Siarhei Siamashka wrote:
On Thu, 5 Feb 2015 22:49:39 -0700 Simon Glass sjg@chromium.org wrote:
Hi Siarhei,
On 5 February 2015 at 02:36, Siarhei Siamashka siarhei.siamashka@gmail.com wrote:
!!! Works only on Allwinner A20 so far and needs a bit more !!! debugging. And even on A20, the transition from the SPL to !!! the main U-boot binary has some glitches.
Now it should be possible to load and execute the same U-Boot SPL, as used for booting from SD cards. The only argument for this new command is the name of the SPL binary file (with eGON header). It can be run as:
fel spl u-boot-sunxi-with-spl.binThis looks great! Does it write only the SPL portion or does it write U-Boot also?
Yes, it only takes care of the SPL part at the moment, but surely this can be improved later.
The biggest inconvenience is that we need to provide the 'fel' tool with with the magic address for the main u-boot binary. This address is hidden in the u-boot sources:
http://git.denx.de/?p=u-boot.git;a=blob;f=include/configs/sunxi-common.h;h=6cfd7e148900#l28But the users need to know it in order to set in as the command line argument for the 'fel' tool.
We could probably do a better job and store the necessary addresses in the SPL header. Currently 'mksunxiboot' provides only a minimalistic header with the "eGON.BT0" signature and a checksum, because otherwise it will refuse to boot from the SD card:
http://git.denx.de/?p=u-boot.git;a=blob;f=tools/mksunxiboot.cBut probably we can extend this header and also add the CONFIG_SYS_TEXT_BASE value there. Together with "kernel_addr_r", "fdt_addr_r", "scriptaddr" and "ramdisk_addr_r" values. After this change, the 'fel' tool will be able to extract all the necessary information from u-boot-sunxi-with-spl.bin file without relying on the command line too much. Moreover, these addresses are sometimes changing, and some users are IMHO rightfully upset whenever this happens:
I think that adding a new header for this sounds like a good idea, this header should start with some 8 byte magic marker so that the fel tool can be 100% sure if it is there.
http://permalink.gmane.org/gmane.comp.hardware.netbook.arm.sunxi/14919I don't really understand how it works but will leave that to you and Hans.
OK :) Anyway, now I have finally pushed a supposedly properly working implementation to the following github branch:
https://github.com/ssvb/sunxi-tools/commits/20150206-fel-large-spl-support
Cool (note I've not looked at the actual code yet), this all sounds quite nice.
It would be really cool if when the new header is present and with a new enough fel tool + u-boot bin we could do:
fel spl u-boot-sunxi-with-spl.bin zImage - dtb
or:
fel spl u-boot-sunxi-with-spl.bin zImage initrd dtb
And everything would just work, note that the "zImage initrd dtb" order is the same as u-boot's bootm command, and the - is there to indicate there is no initrd.
Maybe we should also do something for old kernels where script.bin needs to be loaded, maybe:
fel spl-old-kernel ... ??
Yes, we can handle most of this stuff with Hans. But some communication with you is still needed. In particular, we still need to decide a way to distinguish between booting from the SD card and over USB. Your RFC/WIP patch from http://lists.denx.de/pipermail/u-boot/2015-February/203858.html was checking the value in the 'lr' register, but this is not going to work anymore. To address this, right now I'm just replacing the "eGON.BT0" signature in memory to a new "eGON.FEL" signature at the address 0x0004. The idea is that this signature is critical for booting from the SD card. If there is some other value instead, then we are pretty sure that the SPL was not booted from the SD card. And having it as "eGON.FEL" serves as an indication of what exactly it is.
I think that replacing eGON.BT0 with eGON.FEL when loading the SPL through the new mode is a good way of doing this, ACK to that.
Additionally, we might even not need to store the "lr" and "sp" registers in save_boot_params(). If we settle on using the "eGON.FEL" signature in memory for distinguishing between the FEL and the SD boot, then we can also reuse the other parts of of header (for example, the checksum) to store the return address.
We'll have to see how this all works out to determine whether or not that is a good idea.
As you have figured out earlier, avoiding setting VBAR and CP15 stuff is important. Right now the FEL code in the BROM has an active interrupt handler in use (which I'm suppressing by disabling interrupts early) and the SPL needs to keep interrupts disabled, while trying not to mess its configuration.
Ack.
Regards,
Hans