Re: [PATCH] efi/x86-mixed: leave RET unmitigated but move it into .rodata

[Borislav Petkov <bp@suse.de>]

On Sun, Jul 24, 2022 at 08:34:36PM +0200, Ard Biesheuvel wrote:
> Are you sure you fixed up the conflict correctly? It seems the
> __efi64_thunk end marker ends up in .rodata in your case.

Yep, I f*cked up the merge even if it was pretty easy in meld - sorry
about that.

Now it is correct but it complains differently:

vmlinux.o: warning: objtool: efi_thunk_query_variable_info_nonblocking+0x1ba: unreachable instruction

$ ./scripts/faddr2line vmlinux.o efi_thunk_query_variable_info_nonblocking+0x1ba
efi_thunk_query_variable_info_nonblocking+0x1ba/0x330:
efi_thunk_query_variable_info_nonblocking at /home/boris/kernel/linux/arch/x86/platform/efi/efi_64.c:787
(inlined by) efi_thunk_query_variable_info_nonblocking at /home/boris/kernel/linux/arch/x86/platform/efi/efi_64.c:769

and looking at the asm, it points to:

# 0 "" 2
#NO_APP
	movq	efi(%rip), %rax	# efi.runtime, efi.runtime
	movl	12(%rsp), %r8d	# %sfp, prephitmp_87
	leaq	16(%rsp), %r9	#,
	movl	%r15d, %ecx	# _104,
	movl	%r14d, %edx	# _95,
	movl	%ebp, %esi	# attr,
	movl	76(%rax), %edi	# _30->mixed_mode.query_variable_info, _30->mixed_mode.query_variable_info
	call	__efi64_thunk	#
#APP
# 787 "arch/x86/platform/efi/efi_64.c" 1

1:	movl %r12d,%ds			# __val		<---

this here, after the __efi64_thunk call, which is that segment restoring
after the __efi_thunk call:

	loadsegment(ds, __ds);

Weird, I don't see why though - that should be reachable.

-- 
Regards/Gruss,
    Boris.

SUSE Software Solutions Germany GmbH
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Martje Boudien Moerman
(HRB 36809, AG Nürnberg)