While investigating a FEL boot failure on the OrangePi Zero 2 board (with an H616 SoC), I stared at our FEL code and found some issues, which this series fixes.
Unfortunately those didn't fix the H616 problem, but they are worth having anyway. For the records on the FEL failure: with certain H616 DRAM parameters, newer GCCs (starting with GCC 11, GCC 10 is fine) generate an SPL binary that doesn't properly return to the BootROM after the SPL code has finished. This does not occur on other H616 boards using different DRAM parameters. Staring at the disassemblies from the different compiler versions for a while didn't show anything obvious, the best theory so far is that it's due to a subtle timing issue in the DRAM initialisation code. Inserting a "udelay(0);" at the beginning of mctl_phy_read_calibration() seems to avoid the problem, but is obviously not a proper fix. I will keep looking. However the issues addressed in this series should be fixed, regardless. Cheers, Andre Andre Przywara (3): sunxi: armv8: fel: load only 32-bit values sunxi: h616: lower SPL stack address to avoid BROM data sunxi: fel: drop redundant "control register" save/restore arch/arm/cpu/armv7/sunxi/fel_utils.S | 4 ---- arch/arm/cpu/armv8/fel_utils.S | 8 ++++---- arch/arm/mach-sunxi/board.c | 1 - include/configs/sunxi-common.h | 2 +- 4 files changed, 5 insertions(+), 10 deletions(-) -- 2.25.1
