TSC (time stamp counter) and CX8 (CMPXCHG8B) hardware support"[0][1]
I don't know what these processor features actually do. Also from [0] it looks like it is also removing the floating point emulation code for CPUs without hardware floating point support e.g. 486-SX
[0] https://lore.kernel.org/lkml/20250425084216.3913608-1-mingo@...
[1] https://www.phoronix.com/news/Linux-RFC-Remove-i486-Early-58...
I'm afraid saying it is missing a feature, but unable to say whether that feature is required for anything is what the gpp is complaining about and is an excuse not to support it.
Now if the combobulator transactions with the injunction function to provide the higher level of precision required, then that could be an argument to deprecate the architecture.
So dropping the i80386 wasn't necessary, and saying it was because of the lack of atomic instructions was an excuse, because the real reason was that they didn't want to keep i80386 atomics around.