Michal Jireš <mji...@suse.cz> writes: > On 9/4/25 8:53 PM, Sam James wrote: >> Sam James <s...@gentoo.org> writes: >> >>> Andi Kleen <a...@linux.intel.com> writes: >>> >>>> Sam James <s...@gentoo.org> writes: >>>> >>>>> Michal Jires <mji...@suse.cz> writes: >>>>> >>>>>> I did handle node->iterate_referring, but forgot cnode->callers. >>>>>> >>>>>> Only change are contents of the newly separated >>>>>> mark_symbol_referenced_from_asm >>>>> >>>>> Thanks, I'll try the new patch now. >>>>> >>>>> With the workaround I mentioned earlier, I managed to build but got this >>>>> when booting: >>>> >>>>> Sep 04 02:04:26 mop kernel: nct6775: Unknown symbol vid_from_reg (err -2) >>>> >>>> Likely you need a __visible to MODULE_EXPORT >>> >>> This one doesn't do it, will look around more: >>> >>> --- a/include/linux/export.h >>> +++ b/include/linux/export.h >>> @@ -73,7 +73,7 @@ >>> #endif >>> >>> #define __EXPORT_SYMBOL(sym, license, ns) \ >>> - extern typeof(sym) sym; \ >>> + extern __visible typeof(sym) sym; \ >>> __ADDRESSABLE(sym) \ >>> __GENDWARFKSYMS_EXPORT(sym) \ >>> asm(__stringify(___EXPORT_SYMBOL(sym, license, ns))) >>> >>> (I tried it on the non-extern part too but then I get loads of >>> -Wattribute warnings.) >>> >>> __ADDRESSABLE does include __used at least. >>> >>> I can't yet spot where else it'd need to be, as everything relevant uses >>> that >>> macro indirectly AFAICT. >> I think it's because of: >> commit ddb5cdbafaaad6b99d7007ae1740403124502d03 >> Author: Masahiro Yamada <masahi...@kernel.org> >> AuthorDate: Mon Jun 12 00:50:52 2023 +0900 >> Commit: Masahiro Yamada <masahi...@kernel.org> >> CommitDate: Thu Jun 22 21:17:10 2023 +0900 >> kbuild: generate KSYMTAB entries by modpost >> >>> >>>> >>>> -Andi > > I think the issue is that modpost grabs exported symbol info from > object files which only contain our LTO data. Some of the "SILENCE" > patches disable CRC checks for the same reason. > I didn't try to make modules work yet. > Though it should work if modpost would grab symbol info from vmlinux.o > Or fat-lto might work. > > __visible should not be needed outside of few edge cases. > In this case it is captured in the asm below, so definitely not needed. > > Sidenote: __ADDRESSABLE's implementation guarantees symbol's existence, > but it can be renamed. I don't think it can be implemented correctly > in current GCC (so that it works both inside and outside of functions). > Fortunately it does not seem to be needed to be fixed right now.
Thanks both! sam
