Re: [PATCH -next v4 3/4] arm64: mm: add support for page table check

[Anshuman Khandual <anshuman.khandual@arm.com>]

On 4/18/22 21:50, Pasha Tatashin wrote:
> On Mon, Apr 18, 2022 at 11:47 AM Tong Tiangen <tongtiangen@huawei.com> wrote:
>>
>>
>>
>> 在 2022/4/18 17:28, Anshuman Khandual 写道:
>>> On 4/18/22 09:14, Tong Tiangen wrote:
>>>> From: Kefeng Wang <wangkefeng.wang@huawei.com>
>>>>
>> [...]
>>>>   #endif
>>>
>>> Ran this series on arm64 platform after enabling
>>>
>>> - CONFIG_PAGE_TABLE_CHECK
>>> - CONFIG_PAGE_TABLE_CHECK_ENFORCED (avoiding kernel command line option)
>>>
>>> After some time, the following error came up
>>>
>>> [   23.266013] ------------[ cut here ]------------
>>> [   23.266807] kernel BUG at mm/page_table_check.c:90!
>>> [   23.267609] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
>>> [   23.268503] Modules linked in:
>>> [   23.269012] CPU: 1 PID: 30 Comm: khugepaged Not tainted 5.18.0-rc3-00004-g60aa8e363a91 #2
>>> [   23.270383] Hardware name: linux,dummy-virt (DT)
>>> [   23.271210] pstate: 40400005 (nZcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>>> [   23.272445] pc : page_table_check_clear.isra.6+0x114/0x148
>>> [   23.273429] lr : page_table_check_clear.isra.6+0x64/0x148
>>> [   23.274395] sp : ffff80000afb3ca0
>>> [   23.274994] x29: ffff80000afb3ca0 x28: fffffc00022558e8 x27: ffff80000a27f628
>>> [   23.276260] x26: ffff800009f9f2b0 x25: ffff00008a8d5000 x24: ffff800009f09fa0
>>> [   23.277527] x23: 0000ffff89e00000 x22: ffff800009f09fb8 x21: ffff000089414cc0
>>> [   23.278798] x20: 0000000000000200 x19: fffffc00022a0000 x18: 0000000000000001
>>> [   23.280066] x17: 0000000000000001 x16: 0000000000000000 x15: 0000000000000003
>>> [   23.281331] x14: 0000000000000068 x13: 00000000000000c0 x12: 0000000000000010
>>> [   23.282602] x11: fffffc0002320008 x10: fffffc0002320000 x9 : ffff800009fa1000
>>> [   23.283868] x8 : 00000000ffffffff x7 : 0000000000000001 x6 : ffff800009fa1f08
>>> [   23.285135] x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000
>>> [   23.286406] x2 : 00000000ffffffff x1 : ffff000080f2800c x0 : ffff000080f28000
>>> [   23.287673] Call trace:
>>> [   23.288123]  page_table_check_clear.isra.6+0x114/0x148
>>> [   23.289043]  __page_table_check_pmd_clear+0x3c/0x50
>>> [   23.289918]  pmdp_collapse_flush+0x114/0x370
>>> [   23.290692]  khugepaged+0x1170/0x19e0
>>> [   23.291356]  kthread+0x110/0x120
>>> [   23.291945]  ret_from_fork+0x10/0x20
>>> [   23.292596] Code: 91001041 b8e80024 51000482 36fffd62 (d4210000)
>>> [   23.293678] ---[ end trace 0000000000000000 ]---
>>> [   23.294511] note: khugepaged[30] exited with preempt_count 2
>>>
>>> Looking into file mm/page_table_check.c where this problem occured.
>>>
>>> /*
>>>   * An enty is removed from the page table, decrement the counters for that page
>>>   * verify that it is of correct type and counters do not become negative.
>>>   */
>>> static void page_table_check_clear(struct mm_struct *mm, unsigned long addr,
>>>                                     unsigned long pfn, unsigned long pgcnt)
>>> {
>>>          struct page_ext *page_ext;
>>>          struct page *page;
>>>          unsigned long i;
>>>          bool anon;
>>>
>>>          if (!pfn_valid(pfn))
>>>                  return;
>>>
>>>          page = pfn_to_page(pfn);
>>>          page_ext = lookup_page_ext(page);
>>>          anon = PageAnon(page);
>>>
>>>          for (i = 0; i < pgcnt; i++) {
>>>                  struct page_table_check *ptc = get_page_table_check(page_ext);
>>>
>>>                  if (anon) {
>>>                          BUG_ON(atomic_read(&ptc->file_map_count));
>>>                          BUG_ON(atomic_dec_return(&ptc->anon_map_count) < 0);
>>>                  } else {
>>>                          BUG_ON(atomic_read(&ptc->anon_map_count));
>>>   Triggered here ====>>  BUG_ON(atomic_dec_return(&ptc->file_map_count) < 0);
>>>                  }
>>>                  page_ext = page_ext_next(page_ext);
>>>          }
>>> }
>>>
>>> Could you explain what was expected during pmdp_collapse_flush() which when
>>> failed, triggered this BUG_ON() ? This counter seems to be page table check
>>> specific, could it just go wrong ? I have not looked into the details about
>>> page table check mechanism.
>>>
>>> - Anshuman
>>> .
>>
>> Hi Anshuman:
>>
>> Thanks for your job.
>>
>> Let me briefly explain the principle of page table check(PTC).
>>
>> PTC introduces the following struct for page mapping type count:
>> struct page_table_check {
>>          atomic_t anon_map_count;
>>          atomic_t file_map_count;
>> };
>> This structure can be obtained by "lookup_page_ext(page)"
>>
>> When page table entries are set(pud/pmd/pte), page_table_check_set()  is
>> called to increase the page mapping count, Also check for errors (eg:if
>> a page is used for anonymous mapping, then the page cannot be used for
>> file mapping at the same time).
>>
>> When page table entries are clear(pud/pmd/pte), page_table_check_clear()
>>   is called to decrease the page mapping count, Also check for errors.
>>
>> The error check rules are described in the following documents:
>> Documentation/vm/page_table_check.rst
>>
>> The setting and clearing of page table entries are symmetrical.
>>
>> Here __page_table_check_pmd_clear() trigger BUGON which indicates that
>> the pmd entry file mapping count has become negative.
>>
>> I guess if PTC didn't detect this exception, would there have been any
>> problems?
> 
> It is hard to tell what sort of problem has been detected. More
> debugging is needed in order to understand it. A huge file entry is
> being removed from the page table. However, at least one sub page of
> that entry does not have a record that it was added as a file entry to

I guess PMD splitting scenarios should also be taken care as sub pages
will also go via appropriate XXX_set_at() helpers ?

> the page table. At Google we found a few internal security bugs using
> PTCs. However, this being new on ARM64, it is possible that the bug is
> in PTC/khugepaged itself.
> 
> Anshuman is it possible to repro your scenario in QEMU?

I have been unable to reproduce this reported problem. Last time it just
happened after a fresh boot without anything in particular running. Will
continue experimenting.