Re: Re: [PATCH] mm/memory-failure: don't allow to unpoison hw corrupted page

[zhenwei pi]

On 6/8/22 07:43, HORIGUCHI NAOYA(堀口 直也) wrote:
> On Tue, Jun 07, 2022 at 02:59:59PM -0700, Andrew Morton wrote:
> > On Tue, 7 Jun 2022 14:36:00 +0200 David Hildenbrand <david@xxxxxxxxxx> wrote:
> >
> > > On 06.06.22 11:15, HORIGUCHI NAOYA(堀口 直也) wrote:
> > > > > >     [  917.864266]  <TASK>
> > > > > >     [  917.864961]  clear_huge_page+0x147/0x270
> > > > > >     [  917.866236]  hugetlb_fault+0x440/0xad0
> > > > > >     [  917.867366]  handle_mm_fault+0x270/0x290
> > > > > >     [  917.868532]  do_user_addr_fault+0x1c3/0x680
> > > > > >     [  917.869768]  exc_page_fault+0x6c/0x160
> > > > > >     [  917.870912]  ? asm_exc_page_fault+0x8/0x30
> > > > > >     [  917.872082]  asm_exc_page_fault+0x1e/0x30
> > > > > >     [  917.873220] RIP: 0033:0x7f2aeb8ba367
> > > > > >
> > > > > > I don't think of a workaround for this now ...
> > > > >
> > > > > Could you please tell me how to reproduce this issue?
> > > >
> > > > You are familiar with qemu-monitor-command, so the following procedure
> > > > should work for you:
> > > >
> > > >    - run a process using hugepages on your VM,
> > > >    - check the guest physical address of the hugepage (page-types.c is helpful for this),
> > > >    - inject a MCE with virsh qemu-monitor-command on the guest physical address, then
> > > >    - unpoison the injected physical address.
> > >
> > > That's triggered via debugfs / HWPOISON_INJECT, right?
> > >
> > > That's a DEBUG_KERNEL option, so I'm not 100% sure if we really want to
> > > cc stable.
>
> Sure, the impact of the bug is limited.
>
> > Sure, it's hardly a must-have.  But let's also take the patch
> > complexity&risk into account.  This is one dang simple patch.
> >
> > Or is it.  Should these things be happening outside mf_mutex?  What the
> > heck is the role of mf_mutex anyway?
>
> mf_mutex is to ensure that only one error handling thread can handle
> the pfn at one time, but set_mce_nospec() is called outside it now.
> So if we want to prevent the race with unmap, both of set_mce_nospec()
> and the new kpte check might need to be done in mf_mutex.
>
> - Naoya Horiguchi

OK, I'll sent a v2 patch which includes:
- this change gets protected by mf_mutex
- use -EOPNOTSUPP instead of -EPERM

By the way, I assume that the similar trace(provided by Naoya) is not a 
same issue. It seems undissolved huge page with corrupted KPTE. I'm 
trying to fix it ...

--
zhenwei pi