09:10 <cyrozap> apritzel: Do you mean the address of the fel_stash struct in .data? Or the address of the fel_stash_addr label where the address of the fel_stash struct in .data gets written to? If you meant the former, no, the address of the fel_stash struct in .data hasn't changed (0x28030 in both the working SPL and broken SPL). If you meant the latter, yes, the fel_stash_addr label is located at 0x21ff0 in the

09:10 <cyrozap> working SPL and 0x22060 in the broken one.

09:14 <apritzel> cyrozap: yeah, thanks, was just clasping for straws here

09:15 <apritzel> I started comparing the disassembly of dram_sun50i_h616.o line by line, so far I didn't see any issue (but I am only half-way in)

09:16 <apritzel> a colleague suggested "compiler bisecting": combine half of the .o files from GCC 10 with the other half compiled by GCC 12, then keep narrowing this in

09:16 <apritzel> but it sounds somewhat involved to pull this off in U-Boot

09:25 <cyrozap> apritzel: I did my binary comparison using Ghidra (FOSS) to disassemble the files and Google BinDiff (unfortunately proprietary, but free of charge) to do the comparison, and it saved quite a lot of time for me. If you're not opposed to running the occasional proprietary tool, BinDiff could significantly speed up your diffing efforts since it can compare not just the disassembly, but also the CFGs of

09:25 <cyrozap> the functions.

09:26 <gamiee> oh, didn't knew about bindiff, thanks!

09:28 <cyrozap> Here's the link to download BinDiff, if you're interested: https://www.zynamics.com/software.html

09:33 <cyrozap> The process to use it is: 1) Build BinExport (https://github.com/google/binexport/tree/67be7ce3b30076374c31789636558a7e8b1374fc/java) for your version of Ghidra and install the plugin in Ghidra, 2) Use Ghidra to disassemble each binary, 3) Use Ghidra's export function to export the disassembled binaries as BinExport files, 4) Load the BinExport files into BinDiff for diffing.

09:44 <apritzel> cyrozap: so does that sync back when a few instructions differ, then become the same again? Because that's the main issue when diff'ing objdumps

09:45 <apritzel> for instance I saw that the one GCC version decided to flip the branches of an "if" statement (handling else/if vs. if/else)