Re: [PATCH v2] mm/memory-failure: fix crash in split_huge_page_to_list from soft_offline_page

[Matthew Wilcox]

On Thu, Jan 25, 2024 at 07:53:25PM +0800, Miaohe Lin wrote:
> On 2024/1/24 21:15, Matthew Wilcox wrote:
> >>   Call Trace:
> >>    <TASK>
> >>    ? die+0x32/0x90
> >>    ? do_trap+0xde/0x110
> >>    ? folio_memcg+0xaf/0xd0
> >>    ? do_error_trap+0x60/0x80
> >>    ? folio_memcg+0xaf/0xd0
> >>    ? exc_invalid_op+0x53/0x70
> >>    ? folio_memcg+0xaf/0xd0
> >>    ? asm_exc_invalid_op+0x1a/0x20
> >>    ? folio_memcg+0xaf/0xd0
> >>    ? folio_memcg+0xae/0xd0
> > 
> > I might trim these ? lines out of the backtrace ...
> 
> Do you mean make backtrace looks like something below?
> 
> Call Trace:
>  <TASK>
>  split_huge_page_to_list+0x4d/0x1380
>  ? sysvec_apic_timer_interrupt+0xf/0x80
>  try_to_split_thp_page+0x3a/0xf0
>  soft_offline_page+0x1ea/0x8a0
>  soft_offline_page_store+0x52/0x90
>  kernfs_fop_write_iter+0x118/0x1b0
>  vfs_write+0x30b/0x430
>  ksys_write+0x5e/0xe0
>  do_syscall_64+0xb0/0x1b0
>  entry_SYSCALL_64_after_hwframe+0x6d/0x75
> RIP: 0033:0x7f6c60d14697

Yes.  I'd trim the sysvec_apic_timer_interrupt+0xf/0x80 line too.
These lines aren't actually part of the call trace.  They're addresses
that the unwinder found on the stack but don't actually fit the call
trace.  It puts them in in case they're helpful, but marks them with a ?
to indicate that they're probably not part of the call trace.