On Fri, Feb 28, 2025 at 6:26 PM Peter Zijlstra [off-list ref] wrote:

On Fri, Feb 28, 2025 at 05:53:07PM +0800, Menglong Dong wrote:

quoted

I tested it a little by enabling CFI_CLANG and the extra 5-bytes
padding. It works fine, as mostly CFI_CLANG use
CONFIG_FUNCTION_PADDING_BYTES to find the tags. I'll
do more testing on CFI_CLANG to make sure everything goes
well.

I don't think you understand; please read:

arch/x86/kernel/alternative.c:__apply_fineibt()

and all the code involved with patching FineIBT. I think you'll find it
very broken if you change anything here.

Can you post an actual function preamble from a kernel with
CONFIG_FINEIBT=y with your changes on?

Ex.

$ objdump -wdr build/kernel/futex/core.o

Disassembly of section .text:

0000000000000000 <__cfi_futex_hash>:
       0:       b9 93 0c f9 ad          mov    $0xadf90c93,%ecx

0000000000000005 <.Ltmp0>:
       5:       90                      nop
       6:       90                      nop
       7:       90                      nop
       8:       90                      nop
       9:       90                      nop
       a:       90                      nop
       b:       90                      nop
       c:       90                      nop
       d:       90                      nop
       e:       90                      nop
       f:       90                      nop

0000000000000010 <futex_hash>:
      10:       f3 0f 1e fa             endbr64
      14:       e8 00 00 00 00          call   19 <futex_hash+0x9>      15: R_X86_64_PLT32      __fentry__-0x4
      19:       8b 47 10                mov    0x10(%rdi),%eax


Any change to the layout here *WILL* break the FineIBT code.


If you want to test, make sure your build has FINEIBT=y and boot on an
Intel CPU that has CET-IBT (alderlake and later).

Yeah, I understand now. As I see the definition of CFI_PRE_PADDING, I
know that things is not as simple as I thought, as the layout can be different
with different config. I'll dig it deeper on this part.

Thanks a lot for clearing that up for me. I'll test the CFI_CLANG and
FINEIBT together with this function later.

Thanks!
Menglong Dong