On 06.03.24 09:38, Peng Zhang wrote:
> From: ZhangPeng <[email protected]>
>
> 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 to
> reduce the performance impact in common scenarios.
>
> In our product environment, there were 7 major faults every 12 hours.
> After the patch is applied, no major fault have been triggered.
>
> Testing file 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. 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: 0.22% 0.26% 1.21% -0.15%
> ext4 private file read: 0.03% 1.00% 1.39% 0.34%
> ext4 shared file write: -0.50% -0.02% -0.14% -0.02%
> ramdisk private file write: 0.07% 0.02% 0.53% 0.04%
> ramdisk private file read: 0.01% 1.60% -0.32% -0.02%
>
> [1] https://lore.kernel.org/linux-mm/[email protected]/
> [2] https://github.com/antonblanchard/will-it-scale/
>
> Suggested-by: "Huang, Ying" <[email protected]>
> Suggested-by: David Hildenbrand <[email protected]>
> Signed-off-by: ZhangPeng <[email protected]>
> Signed-off-by: Kefeng Wang <[email protected]>
> ---
> v3->v4:
> - Update the performance data and commit message
> - Check PTE without lock firstly per Huang, Ying
> - Update comments for recheck function per David Hildenbrand
> - Simply return 0 to make it easier to read per David Hildenbrand
> - Check !FAULT_FLAG_ORIG_PTE_VALID instead of pmd_none()
>
> 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 | 46 ++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 46 insertions(+)
>
> diff --git a/mm/filemap.c b/mm/filemap.c
> index b4858d89f1b1..31ab455c4537 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -3181,6 +3181,48 @@ static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
> return fpin;
> }
>
> +static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
> +{
> + struct vm_area_struct *vma = vmf->vma;
> + vm_fault_t ret = 0;
> + pte_t *ptep;
> +
> + /*
> + * We might have COW'ed a pagecache folio and might now have an mlocked
> + * anon folio mapped. The original pagecache folio is not mlocked and
> + * might have been evicted. During a read+clear/modify/write update of
> + * the PTE, such as done in do_numa_page()/change_pte_range(), we
> + * temporarily clear the PTE under PT lock and might detect it here as
> + * "none" when not holding the PT lock.
> + *
> + * Not rechecking the PTE under PT lock could result in an unexpected
> + * major fault in an mlock'ed region. Recheck only for this special
> + * scenario while holding the PT lock, to not degrade non-mlocked
> + * scenarios. Recheck the PTE without PT lock firstly, thereby reducing
> + * the number of times we hold PT lock.
> + */
> + if (!(vma->vm_flags & VM_LOCKED))
> + return 0;
> +
> + if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
> + return 0;
> +
> + ptep = pte_offset_map(vmf->pmd, vmf->address);
> + if (unlikely(!ptep))
> + return VM_FAULT_NOPAGE;
> +
> + if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
> + ret = VM_FAULT_NOPAGE;
> + } else {
> + spin_lock(vmf->ptl);
I assume the assumption is that vmf->ptl is still set from
handle_pte_fault(). But in the meantime, we did a pte_unmap(vmf->pte).
Looking at other users of vmf->ptl, this here sticks out.
I would suggest replacing the pte_offset_map() above by a
pte_offset_map_nolock() that gives us the ptl.
With that
Reviewed-by: David Hildenbrand <[email protected]>
--
Cheers,
David / dhildenb
On 2024/3/6 18:24, David Hildenbrand wrote:
> On 06.03.24 09:38, Peng Zhang wrote:
>> From: ZhangPeng <[email protected]>
>>
>> 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 to
>> reduce the performance impact in common scenarios.
>>
>> In our product environment, there were 7 major faults every 12 hours.
>> After the patch is applied, no major fault have been triggered.
>>
>> Testing file 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. 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: 0.22% 0.26% 1.21% -0.15%
>> ext4 private file read: 0.03% 1.00% 1.39% 0.34%
>> ext4 shared file write: -0.50% -0.02% -0.14% -0.02%
>> ramdisk private file write: 0.07% 0.02% 0.53% 0.04%
>> ramdisk private file read: 0.01% 1.60% -0.32% -0.02%
>>
>> [1]
>> https://lore.kernel.org/linux-mm/[email protected]/
>> [2] https://github.com/antonblanchard/will-it-scale/
>>
>> Suggested-by: "Huang, Ying" <[email protected]>
>> Suggested-by: David Hildenbrand <[email protected]>
>> Signed-off-by: ZhangPeng <[email protected]>
>> Signed-off-by: Kefeng Wang <[email protected]>
>> ---
>> v3->v4:
>> - Update the performance data and commit message
>> - Check PTE without lock firstly per Huang, Ying
>> - Update comments for recheck function per David Hildenbrand
>> - Simply return 0 to make it easier to read per David Hildenbrand
>> - Check !FAULT_FLAG_ORIG_PTE_VALID instead of pmd_none()
>>
>> 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 | 46 ++++++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 46 insertions(+)
>>
>> diff --git a/mm/filemap.c b/mm/filemap.c
>> index b4858d89f1b1..31ab455c4537 100644
>> --- a/mm/filemap.c
>> +++ b/mm/filemap.c
>> @@ -3181,6 +3181,48 @@ static struct file
>> *do_async_mmap_readahead(struct vm_fault *vmf,
>> return fpin;
>> }
>> +static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault
>> *vmf)
>> +{
>> + struct vm_area_struct *vma = vmf->vma;
>> + vm_fault_t ret = 0;
>> + pte_t *ptep;
>> +
>> + /*
>> + * We might have COW'ed a pagecache folio and might now have an
>> mlocked
>> + * anon folio mapped. The original pagecache folio is not
>> mlocked and
>> + * might have been evicted. During a read+clear/modify/write
>> update of
>> + * the PTE, such as done in do_numa_page()/change_pte_range(), we
>> + * temporarily clear the PTE under PT lock and might detect it
>> here as
>> + * "none" when not holding the PT lock.
>> + *
>> + * Not rechecking the PTE under PT lock could result in an
>> unexpected
>> + * major fault in an mlock'ed region. Recheck only for this special
>> + * scenario while holding the PT lock, to not degrade non-mlocked
>> + * scenarios. Recheck the PTE without PT lock firstly, thereby
>> reducing
>> + * the number of times we hold PT lock.
>> + */
>> + if (!(vma->vm_flags & VM_LOCKED))
>> + return 0;
>> +
>> + if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
>> + return 0;
>> +
>> + ptep = pte_offset_map(vmf->pmd, vmf->address);
>> + if (unlikely(!ptep))
>> + return VM_FAULT_NOPAGE;
>> +
>> + if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
>> + ret = VM_FAULT_NOPAGE;
>> + } else {
>> + spin_lock(vmf->ptl);
>
> I assume the assumption is that vmf->ptl is still set from
> handle_pte_fault(). But in the meantime, we did a pte_unmap(vmf->pte).
>
> Looking at other users of vmf->ptl, this here sticks out.
>
> I would suggest replacing the pte_offset_map() above by a
> pte_offset_map_nolock() that gives us the ptl.
>
>
I'll replace it in the next version.
Thanks!
> With that
>
> Reviewed-by: David Hildenbrand <[email protected]>
>
--
Best Regards,
Peng