Re: [PATCH] mm: khugepaged: fix call hpage_collapse_scan_file() for anonymous vma

[Yang Shi <yang@os.amperecomputing.com>]

On 1/10/25 7:54 PM, Liu Shixin wrote:
>
> On 2025/1/11 3:40, Yang Shi wrote:
>>
>>
>> On 1/10/25 11:01 AM, Matthew Wilcox wrote:
>>> On Fri, Jan 10, 2025 at 10:04:42AM -0800, Yang Shi wrote:
>>>> On 1/9/25 8:31 PM, Matthew Wilcox wrote:
>>>>> On Thu, Jan 09, 2025 at 09:00:24AM -0800, Yang Shi wrote:
>>>>>> Thanks for catching this. It sounds a little bit weird to have vm_file for
>>>>>> an anonymous VMA. I'm not sure why we should keep such special case. It
>>>>>> seems shared mapping is treated as shmem file mapping. So can we set vm_file
>>>>>> to NULL when mmap'ing /dev/zero for private mapping? Something like:
>>>>>>
>>>>>> diff --git a/drivers/char/mem.c b/drivers/char/mem.c
>>>>>> index 169eed162a7f..fc332efc5c11 100644
>>>>>> --- a/drivers/char/mem.c
>>>>>> +++ b/drivers/char/mem.c
>>>>>> @@ -527,6 +527,7 @@ static int mmap_zero(struct file *file, struct
>>>>>> vm_area_struct *vma)
>>>>>>            if (vma->vm_flags & VM_SHARED)
>>>>>>                    return shmem_zero_setup(vma);
>>>>>>            vma_set_anonymous(vma);
>>>>>> +       vma->vm_file = NULL;
>>>>>>            return 0;
>>>>>>     }
>>>>> I'm wary this might cause other bugs somewhere.  rc6 is a bit late to be
>>>>> introducing such a subtle change.
>>>> Thanks for the extra caution. Applying the proposed fix in khugepaged code
>>>> is fine to me either. We can try to kill the special case later.
>>>>
>>>> Looking at the code further, I think we should do more to make private
>>>> /dev/zero mapping an anonymous mapping:
>>> I'm still nervous about this.  We map device inodes in a lot of places.
>> Yes, we do. But I don't think this change actually changes the semantic of /dev/zero. Shared /dev/zero mapping is still treated as shmem mapping, private /dev/zero mapping is treated as anonymous mapping, but the current implementation is actually half-baked. It has NULL vma->vm_ops which is used to tell kernel whether it is an anonymous vma or not, but it also has valid vma->vm_file and vma->vm_pgoff as in file offset.
>>
>> So this special case makes kernel has 3 types of VMA:
>>      - anonymous VMA: vm_ops is NULL, vm_file is NULL, vm_pgoff is the linear address pgoff
>>      - file VMA: vm_ops is *NOT* NULL, valid vm_file and vm_pgoff is index in file
>>      - private /dev/zero mapping VMA
>>
> I have posted v2 to fix it in a safe way. Link: https://lore.kernel.org/all/20250111034511.2223353-1-liushixin2@huawei.com/
>
> Maybe we can also revisit commit bfd40eaff5ab ("mm: fix vma_is_anonymous() false-positives") and fix it by another way?

What do you mean about revisiting this commit? Reset vm_file and 
recalculate vm_pgoff in vma_set_anonymous()?

>
> By the way, it seems we collpase the file even after cow for a private file mapping. Is that so?

It seems so.

>
> Thanks,
>> .
>>