Message-ID: <AFA6A945-21AC-4665-A410-E906C7EB3EBF@nvidia.com>
Date: Fri, 26 Apr 2024 16:22:38 -0400
From: Zi Yan <ziy@...dia.com>
To: David Hildenbrand <david@...hat.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
 "Matthew Wilcox (Oracle)" <willy@...radead.org>,
 Yang Shi <shy828301@...il.com>, Ryan Roberts <ryan.roberts@....com>,
 Barry Song <21cnbao@...il.com>, Lance Yang <ioworker0@...il.com>,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5] mm/rmap: do not add fully unmapped large folio to
 deferred split list

On 26 Apr 2024, at 16:15, David Hildenbrand wrote:

> On 26.04.24 21:20, Zi Yan wrote:
>> On 26 Apr 2024, at 15:08, David Hildenbrand wrote:
>>
>>> On 26.04.24 21:02, Zi Yan wrote:
>>>> From: Zi Yan <ziy@...dia.com>
>>>>
>>>> In __folio_remove_rmap(), a large folio is added to deferred split list
>>>> if any page in a folio loses its final mapping. But it is possible that
>>>> the folio is fully unmapped and adding it to deferred split list is
>>>> unnecessary.
>>>>
>>>> For PMD-mapped THPs, that was not really an issue, because removing the
>>>> last PMD mapping in the absence of PTE mappings would not have added the
>>>> folio to the deferred split queue.
>>>>
>>>> However, for PTE-mapped THPs, which are now more prominent due to mTHP,
>>>> they are always added to the deferred split queue. One side effect
>>>> is that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
>>>> unintentionally increased, making it look like there are many partially
>>>> mapped folios -- although the whole folio is fully unmapped stepwise.
>>>>
>>>> Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
>>>> where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
>>>> folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-mapped
>>>> folio is unmapped in one go and can avoid being added to deferred split
>>>> list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still be
>>>> noise when we cannot batch-unmap a complete PTE-mapped folio in one go
>>>> -- or where this type of batching is not implemented yet, e.g., migration.
>>>>
>>>> To avoid the unnecessary addition, folio->_nr_pages_mapped is checked
>>>> to tell if the whole folio is unmapped. If the folio is already on
>>>> deferred split list, it will be skipped, too.
>>>>
>>>> Note: commit 98046944a159 ("mm: huge_memory: add the missing
>>>> folio_test_pmd_mappable() for THP split statistics") tried to exclude
>>>> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not
>>>> fix the above issue. A fully unmapped PTE-mapped order-9 THP was still
>>>> added to deferred split list and counted as THP_DEFERRED_SPLIT_PAGE,
>>>> since nr is 512 (non zero), level is RMAP_LEVEL_PTE, and inside
>>>> deferred_split_folio() the order-9 folio is folio_test_pmd_mappable().
>>>>
>>>> Suggested-by: David Hildenbrand <david@...hat.com>
>>>> Signed-off-by: Zi Yan <ziy@...dia.com>
>>>> ---
>>>>    mm/rmap.c | 12 +++++++++---
>>>>    1 file changed, 9 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/mm/rmap.c b/mm/rmap.c
>>>> index 2608c40dffad..a9bd64ebdd9a 100644
>>>> --- a/mm/rmap.c
>>>> +++ b/mm/rmap.c
>>>> @@ -1495,6 +1495,7 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>>>>    {
>>>>    	atomic_t *mapped = &folio->_nr_pages_mapped;
>>>>    	int last, nr = 0, nr_pmdmapped = 0;
>>>> +	bool partially_mapped = false;
>>>>    	enum node_stat_item idx;
>>>>     	__folio_rmap_sanity_checks(folio, page, nr_pages, level);
>>>> @@ -1515,6 +1516,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>>>>    					nr++;
>>>>    			}
>>>>    		} while (page++, --nr_pages > 0);
>>>> +
>>>> +		partially_mapped = !!nr && !!atomic_read(mapped);
>>>
>>> Nit: The && should remove the need for both !!.
>>
>> My impression was that !! is needed to convert from int to bool and I do
>> find "!!int && !!int" use in the kernel.
>
> I might be wrong about this, but if you wouldn't write
>
> 	if (!!nr && !!atomic_read(mapped))
>
> then
>
> bool partially_mapped = nr && atomic_read(mapped);
>
> is sufficient.
>
> && would make sure that the result is either 0 or 1, which
> you can store safely in a bool, no matter which underlying type
> is used to store that value.
>
> But I *think* nowdays, the compiler will always handle that
> correctly, even without the "&&" (ever since C99 added _Bool).
>
> Likely, also
>
> 	bool partially_mapped = nr & atomic_read(mapped);
>
> Would nowadays work, but looks stupid.
>
>
> Related: https://lkml.org/lkml/2013/8/31/138
>
> ---
> #include <stdio.h>
> #include <stdbool.h>
> #include <stdint.h>
> #include <inttypes.h>
>
> volatile uint64_t a = 0x8000000000000000ull;
>
> void main (void) {
>         printf("uint64_t a = a: 0x%" PRIx64 "\n", a);
>
>         int i1 = a;
>         printf("int i1 = a: %d\n", i1);
>
>         int i2 = !!a;
>         printf("int i2 = !!a: %d\n", i2);
>
>         bool b1 = a;
>         printf("bool b1 = a: %d\n", b1);
>
>         bool b2 = !!a;
>         printf("bool b2 = !!a: %d\n", b2);
> }
> ---
> $ ./test
> uint64_t a = a: 0x8000000000000000
> int i1 = a: 0
> int i2 = !!a: 1
> bool b1 = a: 1
> bool b2 = !!a: 1
> ---
>
> Note that if bool would be defined as "int", you would need the !!, otherwise you
> would lose information.

Thank you for all above. Really appreciate it! And you are right about && and !!.
My memory on !! must be from the old days and now is refreshed. :)

>
> But even for b1, the gcc generates now:
>
>  40118c:       48 8b 05 7d 2e 00 00    mov    0x2e7d(%rip),%rax        # 404010 <a>
>  401193:       48 85 c0                test   %rax,%rax
>  401196:       0f 95 c0                setne  %al
>
>
> My stdbool.h contains
>
> #define bool	_Bool
>
> And I think C99 added _Bool that makes that work.
>
> But I didn't read the standard, and it's time for the weekend :)

Have a good weekend!

--
Best Regards,
Yan, Zi

Download attachment "signature.asc" of type "application/pgp-signature" (855 bytes)

Powered by blists - more mailing lists