Re: [PATCH v3] filemap: avoid unnecessary major faults in filemap_fault()

["Huang, Ying" <ying.huang@intel.com>]

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 issue[1].
>
> This caused by temporarily cleared PTE during a read+clear/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.
>
> This issue affects our traffic analysis service. The inbound traffic is
> heavy. If a major fault occurs, the I/O schedule is triggered and the
> original I/O is suspended. Generally, the I/O schedule is 0.7 ms. If
> other applications are operating disks, the system needs to wait for
> more than 10 ms. However, the inbound traffic is heavy and the NIC buffer
> is small. As a result, packet loss occurs. But the traffic analysis service
> can't tolerate packet loss.
>
> Fix this by holding PTL and rechecking the PTE in filemap_fault() before
> triggering a major fault. We do this check only if vma is VM_LOCKED. In
> our service test environment, the baseline is 7 major faults / 12 hours.
> After the patch is applied, no major fault will be triggered.
>
> Testing file anonymous page read and write page fault performance in
> ext4, tmpfs 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 indicates some performance regressions.
> We do this check only if vma is VM_LOCKED, therefore, no performance
> regressions is caused for most common cases.
>
> The test results are as follows:
>                           processes processes_idle threads threads_idle
> ext4    private file write: -0.51%    0.08%          -0.03%  -0.04%
> ext4    shared  file write:  0.135%  -0.531%          2.883% -0.772%
> ramdisk private file write: -0.48%    0.23%          -1.08%   0.27%
> ramdisk private file  read:  0.07%   -6.90%          -5.85%  -0.70%

Have you retested with the VM_LOCKED optimization?  Why are there still
performance regression?

> tmpfs   private file write: -0.344%  -0.110%          0.200%  0.145%
> tmpfs   shared  file write:  0.958%   0.101%          2.781% -0.337%
> tmpfs   private file  read: -0.16%    0.00%          -0.12%   0.41%
>
> [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: David Hildenbrand <david@redhat.com>
> Signed-off-by: ZhangPeng <zhangpeng362@huawei.com>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
> v2->v3:
> - Do this check only if vma is VM_LOCKED per David Hildenbrand
> - Hold PTL and recheck the PTE
> - Place the recheck code in a new function filemap_fault_recheck_pte()
>
> v1->v2:
> - Add more test results per Huang, Ying
> - Add more comments before check PTE per Huang, Ying, David Hildenbrand
>   and Yin Fengwei
> - Change pte_offset_map_nolock to pte_offset_map as the PTL won't
>   be used
>
> 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 | 40 ++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 40 insertions(+)
>
> diff --git a/mm/filemap.c b/mm/filemap.c
> index b4858d89f1b1..2668bac68df7 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -3181,6 +3181,42 @@ static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
>  	return fpin;
>  }
>  
> +/*
> + * filemap_fault_recheck_pte - hold PTL and recheck whether pte is none.
> + * @vmf - the vm_fault for this fault.
> + *
> + * Recheck PTE as the PTE can be cleared temporarily during a read+clear/modify
> + * /write update of the PTE, eg, do_numa_page()/change_pte_range(). This will
> + * trigger an unexpected major fault, even if we use mlockall(), which may
> + * increase IO and thus cause other unexpected behavior.
> + *
> + * Return VM_FAULT_NOPAGE if the PTE is not none or pte_offset_map_lock()
> + * fails. In other cases, 0 is returned.
> + */
> +static vm_fault_t filemap_fault_recheck_pte(struct vm_fault *vmf)
> +{
> +	struct vm_area_struct *vma = vmf->vma;
> +	vm_fault_t ret = 0;
> +	pte_t *ptep;
> +
> +	if (!(vma->vm_flags & VM_LOCKED))
> +		return ret;
> +
> +	if (pmd_none(*vmf->pmd))
> +		return ret;
> +

How about check PTE without lock firstly?  I guess that this can improve
performance in common case (no race).

> +	ptep = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
> +				   &vmf->ptl);
> +	if (unlikely(!ptep))
> +		return VM_FAULT_NOPAGE;
> +
> +	if (unlikely(!pte_none(ptep_get(ptep))))
> +		ret = VM_FAULT_NOPAGE;
> +
> +	pte_unmap_unlock(ptep, vmf->ptl);
> +	return ret;
> +}
> +
>  /**
>   * filemap_fault - read in file data for page fault handling
>   * @vmf:	struct vm_fault containing details of the fault
> @@ -3236,6 +3272,10 @@ vm_fault_t filemap_fault(struct vm_fault *vmf)
>  			mapping_locked = true;
>  		}
>  	} else {
> +		ret = filemap_fault_recheck_pte(vmf);
> +		if (unlikely(ret))
> +			return ret;
> +
>  		/* No page in the page cache at all */
>  		count_vm_event(PGMAJFAULT);
>  		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);

--
Best Regards,
Huang, Ying