Re: [PATCH] filemap: avoid unnecessary major faults in filemap_fault()

["zhangpeng (AS)" <zhangpeng362@huawei.com>]

On 2024/2/5 15:31, David Hildenbrand wrote:

> On 05.02.24 08:24, zhangpeng (AS) wrote:
>> On 2024/2/5 14:52, Huang, Ying wrote:
>>
>>> "zhangpeng (AS)" <zhangpeng362@huawei.com> writes:
>>>> On 2024/2/5 10:56, Huang, Ying wrote:
>>>>> Peng Zhang <zhangpeng362@huawei.com> writes:
>>>>>> From: ZhangPeng <zhangpeng362@huawei.com>
>>>>>>
>>>>>> The major fault occurred when using mlockall(MCL_CURRENT | 
>>>>>> MCL_FUTURE)
>>>>>> in application, which leading to an unexpected performance issue[1].
>>>>>>
>>>>>> This caused by temporarily cleared PTE during a read/modify/write 
>>>>>> update
>>>>>> of the PTE, eg, do_numa_page()/change_pte_range().
>>>>>>
>>>>>> For the data segment of the user-mode program, the global 
>>>>>> variable area
>>>>>> is a private mapping. After the pagecache is loaded, the private 
>>>>>> anonymous
>>>>>> page is generated after the COW is triggered. Mlockall can lock 
>>>>>> COW pages
>>>>>> (anonymous pages), but the original file pages cannot be locked 
>>>>>> and may
>>>>>> be reclaimed. If the global variable (private anon page) is 
>>>>>> accessed when
>>>>>> vmf->pte is zeroed in numa fault, a file page fault will be 
>>>>>> triggered.
>>>>>>
>>>>>> At this time, the original private file page may have been 
>>>>>> reclaimed.
>>>>>> If the page cache is not available at this time, a major fault 
>>>>>> will be
>>>>>> triggered and the file will be read, causing additional overhead.
>>>>>>
>>>>>> Fix this by rechecking the PTE without acquiring PTL in 
>>>>>> filemap_fault()
>>>>>> before triggering a major fault.
>>>>>>
>>>>>> Testing file anonymous page read and write page fault performance 
>>>>>> in ext4
>>>>>> and ramdisk using will-it-scale[2] on a x86 physical machine. The 
>>>>>> data
>>>>>> is the average change compared with the mainline after the patch is
>>>>>> applied. The test results are within the range of fluctuation, 
>>>>>> and there
>>>>>> is no obvious difference. The test results are as follows:
>>>>>>             processes processes_idle threads threads_idle
>>>>>> ext4 file write:    -1.14%    -0.08%         -1.87% 0.13%
>>>>>> ext4 file read:         0.03%      -0.65% -0.51%    -0.08%
>>>>>> ramdisk file write:    -1.21%    -0.21%         -1.12% 0.11%
>>>>>> ramdisk file read:     0.00%    -0.68%         -0.33% -0.02%
>>>>>>
>>>>>> [1] 
>>>>>> https://lore.kernel.org/linux-mm/9e62fd9a-bee0-52bf-50a7-498fa17434ee@huawei.com/
>>>>>> [2] https://github.com/antonblanchard/will-it-scale/
>>>>>>
>>>>>> Suggested-by: "Huang, Ying" <ying.huang@intel.com>
>>>>>> Suggested-by: Yin Fengwei <fengwei.yin@intel.com>
>>>>>> Signed-off-by: ZhangPeng <zhangpeng362@huawei.com>
>>>>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>>>>>> ---
>>>>>> RFC->v1:
>>>>>> - Add error handling when ptep == NULL per Huang, Ying and 
>>>>>> Matthew Wilcox
>>>>>> - Check the PTE without acquiring PTL in filemap_fault(), 
>>>>>> suggested by
>>>>>>      Huang, Ying and Yin Fengwei
>>>>>> - Add pmd_none() check before PTE map
>>>>>> - Update commit message and add performance test information
>>>>>>
>>>>>>     mm/filemap.c | 18 ++++++++++++++++++
>>>>>>     1 file changed, 18 insertions(+)
>>>>>>
>>>>>> diff --git a/mm/filemap.c b/mm/filemap.c
>>>>>> index 142864338ca4..b29cdeb6a03b 100644
>>>>>> --- a/mm/filemap.c
>>>>>> +++ b/mm/filemap.c
>>>>>> @@ -3238,6 +3238,24 @@ vm_fault_t filemap_fault(struct vm_fault 
>>>>>> *vmf)
>>>>>>                 mapping_locked = true;
>>>>>>             }
>>>>>>         } else {
>>>>>> +        if (!pmd_none(*vmf->pmd)) {
>>>>>> +            pte_t *ptep;
>>>>>> +
>>>>>> +            ptep = pte_offset_map_nolock(vmf->vma->vm_mm, vmf->pmd,
>>>>>> +                             vmf->address, &vmf->ptl);
>>>>>> +            if (unlikely(!ptep))
>>>>>> +                return VM_FAULT_NOPAGE;
>>>>>> +            /*
>>>>>> +             * Recheck pte as the pte can be cleared temporarily
>>>>>> +             * during a read/modify/write update.
>>>>>> +             */
>>>>> I think that we should add some comments here about the racy 
>>>>> checking.
>>>> I'll add comments in a v2 as follows:
>>>> /*
>>>>    * Recheck PTE as the PTE can be cleared temporarily
>>>>    * during a read/modify/write update of the PTE, eg,
>>>>    * do_numa_page()/change_pte_range(). This will trigger
>>>>    * a major fault, even if we use mlockall, which may
>>>>    * affect performance.
>>>>    */
>>> Sorry, my previous words aren't clear enough.  I mean some comments as
>>> follows,
>>>
>>> We don't hold PTL here, so the check is still racy.  But acquiring PTL
>>> hurts performance and the race window seems small enough.
>>
>> Got it. I'll add comments in a v2 as follows:
>> /*
>>    * Recheck PTE as the PTE can be cleared temporarily
>>    * during a read/modify/write update of the PTE.
>>    * We don't hold PTL here as acquiring PTL hurts
>>    * performance. So the check is still racy, but
>>    * the race window seems small enough.
>>    */
>
> It'd be worth spelling out what happens when we lose the race.
>
I'll add what happens when we lose the race as follows:
/*
  * Recheck PTE as the PTE can be cleared temporarily
  * during a read/modify/write update of the PTE, eg,
  * do_numa_page()/change_pte_range(). This will trigger
  * a major fault, even if we use mlockall, which may
  * affect performance.
  * We don't hold PTL here as acquiring PTL hurts
  * performance. So the check is still racy, but
  * the race window seems small enough.
  */

-- 
Best Regards,
Peng