Re: [linus:master] [x86/bugs] 6613d82e61: general_protection_fault:#[##]

From: Linux regression tracking (Thorsten Leemhuis)
Date: Sun Apr 14 2024 - 02:42:10 EST


Hi, Thorsten here, the Linux kernel's regression tracker.

On 28.03.24 22:17, Pawan Gupta wrote:
> On Thu, Mar 28, 2024 at 03:36:28PM +0800, kernel test robot wrote:
>> compiler: clang-17
>> test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
>>
>> If you fix the issue in a separate patch/commit (i.e. not just a new version of
>> the same patch/commit), kindly add following tags
>> | Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
>> | Closes: https://lore.kernel.org/oe-lkp/202403281553.79f5a16f-lkp@xxxxxxxxx

TWIMC, a user report general protection faults with dosemu that were
bisected to a 6.6.y backport of the commit that causes the problem
discussed in this thread (6613d82e617dd7 ("x86/bugs: Use ALTERNATIVE()
instead of mds_user_clear static key")).

User compiles using gcc, so it might be a different problem. Happens
with 6.8.y as well.

The problem occurs with x86-32 kernels, but strangely only on some of
the x86-32 systems the reporter has (e.g. on some everything works
fine). Makes me wonder if the commit exposed an older problem that only
happens on some machines.

For details see https://bugzilla.kernel.org/show_bug.cgi?id=218707
Could not CC the reporter here due to the bugzilla privacy policy; if
you want to get in contact, please use bugzilla.

Ciao, Thorsten

>> [ 25.175767][ T670] VFS: Warning: trinity-c2 using old stat() call. Recompile your binary.
>> [ 25.245597][ T669] general protection fault: 0000 [#1] PREEMPT SMP
>> [ 25.246417][ T669] CPU: 1 PID: 669 Comm: trinity-c1 Not tainted 6.8.0-rc5-00004-g6613d82e617d #1 85a4928d2e6b42899c3861e57e26bdc646c4c5f9
>> [ 25.247743][ T669] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.2-debian-1.16.2-1 04/01/2014
>> [ 25.248865][ T669] EIP: restore_all_switch_stack (kbuild/src/consumer/arch/x86/entry/entry_32.S:957)
>> [ 25.249510][ T669] Code: 4c 24 10 36 89 48 fc 8b 4c 24 0c 81 e1 ff ff 00 00 36 89 48 f8 8b 4c 24 08 36 89 48 f4 8b 4c 24 04 36 89 48 f0 59 8d 60 f0 58 <0f> 00 2d 00 94 d5 c1 cf 6a 00 68 88 6b d4 c1 eb 00 fc 0f a0 50 b8
>> All code
>> ========
>> 0: 4c 24 10 rex.WR and $0x10,%al
>> 3: 36 89 48 fc ss mov %ecx,-0x4(%rax)
>> 7: 8b 4c 24 0c mov 0xc(%rsp),%ecx
>> b: 81 e1 ff ff 00 00 and $0xffff,%ecx
>> 11: 36 89 48 f8 ss mov %ecx,-0x8(%rax)
>> 15: 8b 4c 24 08 mov 0x8(%rsp),%ecx
>> 19: 36 89 48 f4 ss mov %ecx,-0xc(%rax)
>> 1d: 8b 4c 24 04 mov 0x4(%rsp),%ecx
>> 21: 36 89 48 f0 ss mov %ecx,-0x10(%rax)
>> 25: 59 pop %rcx
>> 26: 8d 60 f0 lea -0x10(%rax),%esp
>> 29: 58 pop %rax
>> 2a:* 0f 00 2d 00 94 d5 c1 verw -0x3e2a6c00(%rip) # 0xffffffffc1d59431 <-- trapping instruction
>
> This is due to 64-bit addressing with CONFIG_X86_32=y on clang.
>
> I haven't tried with clang, but I don't see this happening with gcc-11:
>
> entry_INT80_32:
> ...
> <+446>: mov 0x4(%esp),%ecx
> <+450>: mov %ecx,%ss:-0x10(%eax)
> <+454>: pop %ecx
> <+455>: lea -0x10(%eax),%esp
> <+458>: pop %eax
> <+459>: verw 0xc1d5c700 <----------
> <+466>: iret
>
>> 31: cf iret
>> 32: 6a 00 push $0x0
>> 34: 68 88 6b d4 c1 push $0xffffffffc1d46b88
>> 39: eb 00 jmp 0x3b
> ...
>
> The config has CONFIG_X86_32=y, but it is possible that in 32-bit build
> with clang, 64-bit mode expansion of "VERW (_ASM_RIP(addr))" is getting
> used i.e. __ASM_FORM_RAW(b) below:
>
> file: arch/x86/include/asm/asm.h
> ...
> #ifndef __x86_64__
> /* 32 bit */
> # define __ASM_SEL(a,b) __ASM_FORM(a)
> # define __ASM_SEL_RAW(a,b) __ASM_FORM_RAW(a)
> #else
> /* 64 bit */
> # define __ASM_SEL(a,b) __ASM_FORM(b)
> # define __ASM_SEL_RAW(a,b) __ASM_FORM_RAW(b) <--------
> #endif
> ...
> /* Adds a (%rip) suffix on 64 bits only; for immediate memory references */
> #define _ASM_RIP(x) __ASM_SEL_RAW(x, x (__ASM_REGPFX rip))
>
> Possibly __x86_64__ is being defined with clang even when CONFIG_X86_32=y.
>
> I am not sure about current level of 32-bit mode support in clang. This
> seems inconclusive:
>
> https://discourse.llvm.org/t/x86-32-bit-testing/65480
>
> Does anyone care about 32-bit mode builds with clang?