On Mon, 9 Jun 2025 22:09:34 +0900
Masami Hiramatsu (Google) [off-list ref] wrote:

[...]

Here is the compiled code of _raw_spin_lock.

ffffffff825daa00 <_raw_spin_lock>:
ffffffff825daa00:       f3 0f 1e fa             endbr64
ffffffff825daa04:       e8 47 a6 d5 fe          call   ffffffff81335050 <__fentry__>

Since int3 exception happens after decoded int3 (1 byte), the RIP
`_raw_spin_lock+0x05` is not an instruction boundary.

quoted

<4>[   58.998563] Code: 5d e9 ff 12 00 00 66 66 2e 0f 1f 84 00 00 00
00 00 0f 1f 40 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 f3
0f 1e fa 0f <1f> 44 00 00 55 48 89 e5 53 48 89 fb bf 01 00 00 00 e8 15
12 e4 fe

And the call is already modified back to a 5-bytes nop when we
dump the code. Thus it may hit the intermediate int3 for transforming
code.

e8 47 a6 d5 fe
 (first step)
cc 47 a6 d5 fe
 (second step)
cc 1f 44 00 00 <- hit?
 (third step)
0f 1f 44 00 00 <- handle int3

It is very unlikely scenario (and I'm not sure qemu can correctly
emulate it.) But if a CPU hits the int3 (cc) on _raw_spin_lock()+0x4
before anoter CPU' runs third step in smp_text_poke_batch_finish(),
and before the CPU runs smp_text_poke_int3_handler(), the CPU' runs
the thrid step and sets text_poke_array_refs 0, 
the smp_text_poke_int3_handler() returns 0 and causes the same
problem. 

<CPU0>					<CPU1>
					Start smp_text_poke_batch_finish().
					Finish second step.
Hit int3 (*)
					Finish third step.
					Run smp_text_poke_sync_each_cpu().(**)
					Clear text_poke_array_refs[cpu0]
Start smp_text_poke_int3_handler()
Failed to get text_poke_array_refs[cpu0]
Oops: int3


But as I said it is very unlikely, because as far as I know;

(*) smp_text_poke_int3_handler() is called directly from exc_int3()
   which is a kind of NMI, so other interrupt should not run.
(**) In the third step, smp_text_poke_batch_finish() sends IPI for
   sync core after removing int3. Thus any int3 exception handling
   should be finished.

Maybe one possible scenario is to hit the int3 after the third step
somehow (on I-cache?).

------
<CPU0>					<CPU1>
					Start smp_text_poke_batch_finish().
					Start the third step. (remove INT3)
					on_each_cpu(do_sync_core)
do_sync_core(do SERIALIZE)
					Finish the third step.
Hit INT3 (from I-cache?)
					Clear text_poke_array_refs[cpu0]
Start smp_text_poke_int3_handler()
Failed to get text_poke_array_refs[cpu0]
Oops: int3
------


SERIALIZE instruction may flash pipeline, thus the processor needs
to reload the instruction. But it is not ensured to reload it from
memory because SERIALIZE does not invalidate the cache.

If that hypotheses is correct, we need to invalidate the cache
(flush TLB) in the third step, before the do_sync_core().

Or, if it is unsure, we can just evacuate the kernel from die("int3")
by retrying the new instruction, when the INT3 is disappeared.


Thank you,

-- 
Masami Hiramatsu (Google) [off-list ref]