Re: linux-5.10.11 build failure

[Chris Clayton]

On 28/01/2021 15:52, Josh Poimboeuf wrote:
> On Thu, Jan 28, 2021 at 11:24:47AM +0000, Thomas Backlund wrote:
>> Den 28.1.2021 kl. 12:05, skrev Chris Clayton:
>>>
>>> On 28/01/2021 09:34, Greg Kroah-Hartman wrote:
>>>> On Thu, Jan 28, 2021 at 09:17:10AM +0000, Chris Clayton wrote:
>>>>> Hi,
>>>>>
>>>>> Building 5.10.11 fails on my (x86-64) laptop thusly:
>>>>>
>>>>> ..
>>>>>
>>>>>   AS      arch/x86/entry/thunk_64.o
>>>>>    CC      arch/x86/entry/vsyscall/vsyscall_64.o
>>>>>    AS      arch/x86/realmode/rm/header.o
>>>>>    CC      arch/x86/mm/pat/set_memory.o
>>>>>    CC      arch/x86/events/amd/core.o
>>>>>    CC      arch/x86/kernel/fpu/init.o
>>>>>    CC      arch/x86/entry/vdso/vma.o
>>>>>    CC      kernel/sched/core.o
>>>>> arch/x86/entry/thunk_64.o: warning: objtool: missing symbol for insn at offset 0x3e
>>>>>
>>>>>    AS      arch/x86/realmode/rm/trampoline_64.o
>>>>> make[2]: *** [scripts/Makefile.build:360: arch/x86/entry/thunk_64.o] Error 255
>>>>> make[2]: *** Deleting file 'arch/x86/entry/thunk_64.o'
>>>>> make[2]: *** Waiting for unfinished jobs....
>>>>>
>>>>> ..
>>>>>
>>>>> Compiler is latest snapshot of gcc-10.
>>>>>
>>>>> Happy to test the fix but please cc me as I'm not subscribed
>>>>
>>>> Can you do 'git bisect' to track down the offending commit?
>>>>
>>>
>>> Sure, but I'll hold that request for a while. I updated to binutils-2.36 on Monday and I'm pretty sure that is a feature
>>> of this build fail. I've reverted binutils to 2.35.1, and the build succeeds. Updated to 2.36 again and, surprise,
>>> surprise, the kernel build fails again.
>>>
>>> I've had a glance at the binutils ML and there are all sorts of issues being reported, but it's beyond my knowledge to
>>> assess if this build error is related to any of them.
>>>
>>> I'll stick with binutils-2.35.1 for the time being.
>>>
>>>> And what exact gcc version are you using?
>>>>
>>>
>>>   It's built from the 10-20210123 snapshot tarball.
>>>
>>> I can report this to the binutils folks, but might it be better if the objtool maintainer looks at it first? The
>>> binutils change might just have opened the gate to a bug in objtool.
>>>
>>>> thanks,
>>>>
>>>> greg k-h
>>>>
>>>
>>
>>
>> AFAIK you need this in stable trees:
>>
>>  From 1d489151e9f9d1647110277ff77282fe4d96d09b Mon Sep 17 00:00:00 2001
>> From: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
>> Date: Thu, 14 Jan 2021 16:14:01 -0600
>> Subject: [PATCH] objtool: Don't fail on missing symbol table
> 
> Actually I think you need:
> 
>   5e6dca82bcaa ("x86/entry: Emit a symbol for register restoring thunk")
> 
> I submitted a patch to stable list a few days ago.
> 

Yes, that's what I concluded, Josh. 5.10.11 builds with that patch added but it's not in Linus's tree yet, so, as I
understand it, is not yet a candidate from stable.


> (Though it's possible you need both commits, I'm not sure if binutils
>  2.36 has the symbol stripping stuff)
>