On Mon, Mar 16, 2026 at 10:27:44PM +0100, Arnd Bergmann wrote:
> I've found two more objtool related build failures on today's linux-next.
> 
> See attachment for .config and object files.
> 
> ==> build/x86/0x8F5E6B2F_defconfig/log <==
> mm/memory.o: warning: objtool: folio_zero_user+0x69a: undefined stack state
> mm/memory.o: error: objtool: folio_zero_user+0x69a: unknown CFA base reg -1
> make[5]: *** [/home/arnd/arm-soc/scripts/Makefile.build:279: mm/memory.o] Error 255

This is a clang issue we've seen before, where ASM_CALL_CONSTRAINT
causes clang to save/restore RSP around some inline asm.

The "fix" was going to be asm_call(), as discussed here with Linus:

  https://lore.kernel.org/174099976188.10177.7153571701278544000.tip-bot2@tip-bot2

I need to dust off those patches.

> ==> build/x86/0xEC8A503_defconfig/log <==
> vmlinux.o: warning: objtool: ___bpf_prog_run+0x202: sibling call from callable instruction with modified stack frame
> vmlinux.o: warning: objtool: __x64_sys_get_mempolicy+0x1705: stack state mismatch: cfa1=4+288 cfa2=4+80
> vmlinux.o: error: objtool: vmw_host_printf+0xd: unknown CFA base reg 0
> make[4]: *** [/home/arnd/arm-soc/scripts/Makefile.vmlinux_o:76: vmlinux.o] Error 255

This one's easy, just need to add RAX to the ORC register list.  I will
post a fix shortly.

-- 
Josh