ljs, 1 month ago

@kees ah yeah @vbabka has had some fun experience with 386 well I think gcov stuff actually wasn't it? But probably similar trashing of registers that shouldn't be trashed.

OK so that's probably irrelevant but sort of ballpark :P

Also 32-bit should die

kees, 1 month ago

@ljs @vbabka I think I found it: it looks like the handler calls aren't being emitted with knowledge of -mregparm=3 and are instead pushing arguments to the stack.

kernellogger, 1 month ago

@kees @ljs @vbabka

Clang? 32-bit kernel? That rang a bell in my head, as it reminded me of this:

https://lore.kernel.org/all/202403281553.79f5a16f-lkp@intel.com/

Might be totally unrelated, just thought I mention it in case it might not.