Re: [PATCH] mm: reuse the unshared swapcache page in do_wp_page

[David Hildenbrand <david@redhat.com>]

On 13.01.22 16:02, Matthew Wilcox wrote:
> On Thu, Jan 13, 2022 at 03:46:54PM +0100, David Hildenbrand wrote:
>> On 13.01.22 15:39, Matthew Wilcox wrote:
>>> On Thu, Jan 13, 2022 at 10:03:18PM +0800, Liang Zhang wrote:
>>>> In current implementation, process's read requestions will fault in pages
>>>> with WP flags in PTEs. Next, if process emit a write requestion will go
>>>> into do_wp_page() and copy data to a new allocated page from the old one
>>>> due to refcount > 1 (page table mapped and swapcache), which could be
>>>> result in performance degradation. In fact, this page is exclusively owned
>>>> by this process and the duplication from old to a new allocated page is
>>>> really unnecessary.
>>>>
>>>> So In this situation, these unshared pages can be reused by its process.
>>>
>>> Let's bring Linus in on this, but I think this reintroduces all of the
>>> mapcount problems that we've been discussing recently.
>>>
>>> How about this as an alternative?
>>>
>>> +++ b/mm/memory.c
>>> @@ -3291,11 +3291,11 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
>>>                 struct page *page = vmf->page;
>>>
>>>                 /* PageKsm() doesn't necessarily raise the page refcount */
>>> -               if (PageKsm(page) || page_count(page) != 1)
>>> +               if (PageKsm(page) || page_count(page) != 1 + PageSwapCache(page))
>>>                         goto copy;
>>>                 if (!trylock_page(page))
>>>                         goto copy;
>>> -               if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1) {
>>> +               if (PageKsm(page) || page_mapcount(page) != 1 || page_count(page) != 1 + PageSwapCache(page)) {
>>>                         unlock_page(page);
>>>                         goto copy;
>>>                 }
>>
>> Funny, I was staring at swap reuse code as I received this mail ...
>> because if we're not using reuse_swap_page() here anymore, we shouldn't
>> really be reusing it anywhere for consistency, most prominently in
>> do_swap_page() when we handle vmf->flags & FAULT_FLAG_WRITE just
>> similarly as we do here ...
>>
>> And that's where things get hairy and I am still trying to figure out
>> all of the details.
>>
>> Regarding above: If the page is swapped out in multiple processes but
>> was only faulted into the current process R/O, and then we try to write:
>>
>> 1. Still in the swapcache: PageSwapCache()
>> 2. Mapped only by one process: page_mapcount(page) == 1
>> 3. Reference from one page table and the swap cache: page_count(page) ==
>>
>> But other processes could read-fault on the swapcache page, no?
>>
>> I think we'd really have to check against the swapcount as well ...
>> essentially reuse_swap_page(), no?
> 
> Unfortunately the last digit is missing from your "3.", but I

Sorry, == 2.

> think you're absolutely right; we need to check swapcount.  So
> once reuse_swap_page() checks page_count instead of mapcount, we'll
> be good?
> 

That's something I've been thinking of. Either get rid of
reuse_swap_page() completely or make it obey the same rules everywhere.

It's highly inconsistent how we handle COW.

-- 
Thanks,

David / dhildenb