Re: [PATCH] x86/alternative: fix undefined reference to __ibt_endbr_seal[_end]

[Miaohe Lin]

On 2022/10/9 20:28, Peter Zijlstra wrote:
> On Sun, Oct 09, 2022 at 11:45:32PM +0800, Miaohe Lin wrote:
>> When I compile the kernel bzImage, there are several compile errors:
>>
>> ld: arch/x86/kernel/alternative.o: in function
>> `alternative_instructions':
>> alternative.c:(.init.text+0x15d): undefined reference to
>> `__ibt_endbr_seal_end'
>> ld: alternative.c:(.init.text+0x164): undefined reference to
>> `__ibt_endbr_seal'
>>
>> It's because __ibt_endbr_seal and __ibt_endbr_seal_end are not optimized
>> away when CONFIG_X86_KERNEL_IBT isn't enabled. Remove noinline attribute
>> to help gcc optimize them away.
>>
>> Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
>> ---
>>  arch/x86/kernel/alternative.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
>> index 5cadcea035e0..beaf9fc44e2f 100644
>> --- a/arch/x86/kernel/alternative.c
>> +++ b/arch/x86/kernel/alternative.c
>> @@ -624,7 +624,7 @@ void __init_or_module noinline apply_ibt_endbr(s32 *start, s32 *end)
>>  
>>  #else
>>  
>> -void __init_or_module noinline apply_ibt_endbr(s32 *start, s32 *end) { }
>> +void __init_or_module apply_ibt_endbr(s32 *start, s32 *end) { }
>>  
>>  #endif /* CONFIG_X86_KERNEL_IBT */
>>  
> 
> None of this makes any sense...

This compile error occurs in my env when CONFIG_X86_KERNEL_IBT isn't enabled with gcc version 7.3.0 (GCC).
And the generated assembly code looks like below:

0000000000000104 <alternative_instructions>:
 104:   e8 90 ff ff ff          callq  99 <int3_selftest>
 109:   e8 00 00 00 00          callq  10e <alternative_instructions+0xa>
 10e:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 115:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 11c:   e8 00 00 00 00          callq  121 <alternative_instructions+0x1d>
 121:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 128:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 12f:   e8 00 00 00 00          callq  134 <alternative_instructions+0x30>
 134:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 13b:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 142:   e8 00 00 00 00          callq  147 <alternative_instructions+0x43>
 147:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 14e:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 155:   e8 00 00 00 00          callq  15a <alternative_instructions+0x56>
 >15a:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 >161:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 >168:   e8 00 00 00 00          callq  16d <alternative_instructions+0x69>
 >16d:   83 3d 00 00 00 00 00    cmpl   $0x0,0x0(%rip)        # 174 <alternative_instructions+0x70>
 >174:   75 4a                   jne    1c0 <alternative_instructions+0xbc>

It seems the call to apply_ibt_endbr() is not optimized out when CONFIG_X86_KERNEL_IBT isn't enabled.
And after the patch, the generated assembly code looks like below:

0000000000000104 <alternative_instructions>:
 104:   e8 90 ff ff ff          callq  99 <int3_selftest>
 109:   e8 00 00 00 00          callq  10e <alternative_instructions+0xa>
 10e:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 115:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 11c:   e8 00 00 00 00          callq  121 <alternative_instructions+0x1d>
 121:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 128:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 12f:   e8 00 00 00 00          callq  134 <alternative_instructions+0x30>
 134:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 13b:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 142:   e8 00 00 00 00          callq  147 <alternative_instructions+0x43>
 147:   48 c7 c6 00 00 00 00    mov    $0x0,%rsi
 14e:   48 c7 c7 00 00 00 00    mov    $0x0,%rdi
 155:   e8 00 00 00 00          callq  15a <alternative_instructions+0x56>
 >15a:   83 3d 00 00 00 00 00    cmpl   $0x0,0x0(%rip)        # 161 <alternative_instructions+0x5d>
 >161:   75 4a                   jne    1ad <alternative_instructions+0xa9>

The call to apply_ibt_endbr() is optimized out this time. Maybe it's the problem of gcc. But remove
noinline attribute make compile works in my env. Am I miss something?

Thanks,
Miaohe Lin


> 
> .
>