Message-ID: <760800b1-c708-4c1d-b153-c252ab21938a@redhat.com>
Date: Thu, 28 Mar 2024 13:07:24 +0100
From: David Hildenbrand <david@...hat.com>
To: Baolin Wang <baolin.wang@...ux.alibaba.com>, akpm@...ux-foundation.org
Cc: mgorman@...hsingularity.net, wangkefeng.wang@...wei.com,
 jhubbard@...dia.com, ying.huang@...el.com, 21cnbao@...il.com,
 ryan.roberts@....com, linux-mm@...ck.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/2] mm: support multi-size THP numa balancing

On 28.03.24 12:34, Baolin Wang wrote:
> 
> 
> On 2024/3/28 17:25, David Hildenbrand wrote:
>> On 26.03.24 12:51, Baolin Wang wrote:
>>> Now the anonymous page allocation already supports multi-size THP (mTHP),
>>> but the numa balancing still prohibits mTHP migration even though it
>>> is an
>>> exclusive mapping, which is unreasonable.
>>>
>>> Allow scanning mTHP:
>>> Commit 859d4adc3415 ("mm: numa: do not trap faults on shared data section
>>> pages") skips shared CoW pages' NUMA page migration to avoid shared data
>>> segment migration. In addition, commit 80d47f5de5e3 ("mm: don't try to
>>> NUMA-migrate COW pages that have other uses") change to use page_count()
>>> to avoid GUP pages migration, that will also skip the mTHP numa scaning.
>>> Theoretically, we can use folio_maybe_dma_pinned() to detect the GUP
>>> issue, although there is still a GUP race, the issue seems to have been
>>> resolved by commit 80d47f5de5e3. Meanwhile, use the
>>> folio_likely_mapped_shared()
>>> to skip shared CoW pages though this is not a precise sharers count. To
>>> check if the folio is shared, ideally we want to make sure every page is
>>> mapped to the same process, but doing that seems expensive and using
>>> the estimated mapcount seems can work when running autonuma benchmark.
>>>
>>> Allow migrating mTHP:
>>> As mentioned in the previous thread[1], large folios (including THP) are
>>> more susceptible to false sharing issues among threads than 4K base page,
>>> leading to pages ping-pong back and forth during numa balancing, which is
>>> currently not easy to resolve. Therefore, as a start to support mTHP numa
>>> balancing, we can follow the PMD mapped THP's strategy, that means we can
>>> reuse the 2-stage filter in should_numa_migrate_memory() to check if the
>>> mTHP is being heavily contended among threads (through checking the
>>> CPU id
>>> and pid of the last access) to avoid false sharing at some degree. Thus,
>>> we can restore all PTE maps upon the first hint page fault of a large
>>> folio
>>> to follow the PMD mapped THP's strategy. In the future, we can
>>> continue to
>>> optimize the NUMA balancing algorithm to avoid the false sharing issue
>>> with
>>> large folios as much as possible.
>>>
>>> Performance data:
>>> Machine environment: 2 nodes, 128 cores Intel(R) Xeon(R) Platinum
>>> Base: 2024-03-25 mm-unstable branch
>>> Enable mTHP to run autonuma-benchmark
>>>
>>> mTHP:16K
>>> Base                Patched
>>> numa01                numa01
>>> 224.70                137.23
>>> numa01_THREAD_ALLOC        numa01_THREAD_ALLOC
>>> 118.05                50.57
>>> numa02                numa02
>>> 13.45                9.30
>>> numa02_SMT            numa02_SMT
>>> 14.80                7.43
>>>
>>> mTHP:64K
>>> Base                Patched
>>> numa01                numa01
>>> 216.15                135.20
>>> numa01_THREAD_ALLOC        numa01_THREAD_ALLOC
>>> 115.35                46.93
>>> numa02                numa02
>>> 13.24                9.24
>>> numa02_SMT            numa02_SMT
>>> 14.67                7.31
>>>
>>> mTHP:128K
>>> Base                Patched
>>> numa01                numa01
>>> 205.13                140.41
>>> numa01_THREAD_ALLOC        numa01_THREAD_ALLOC
>>> 112.93                44.78
>>> numa02                numa02
>>> 13.16                9.19
>>> numa02_SMT            numa02_SMT
>>> 14.81                7.39
>>>
>>> [1]
>>> https://lore.kernel.org/all/20231117100745.fnpijbk4xgmals3k@techsingularity.net/
>>> Signed-off-by: Baolin Wang <baolin.wang@...ux.alibaba.com>
>>> ---
>>>    mm/memory.c   | 56 +++++++++++++++++++++++++++++++++++++++++++--------
>>>    mm/mprotect.c |  3 ++-
>>>    2 files changed, 50 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index c30fb4b95e15..36191a9c799c 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -5068,16 +5068,55 @@ static void numa_rebuild_single_mapping(struct
>>> vm_fault *vmf, struct vm_area_str
>>>        update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1);
>>>    }
>>> +static void numa_rebuild_large_mapping(struct vm_fault *vmf, struct
>>> vm_area_struct *vma,
>>> +                       struct folio *folio, pte_t fault_pte, bool
>>> ignore_writable)
>>> +{
>>> +    int nr = pte_pfn(fault_pte) - folio_pfn(folio);
>>> +    unsigned long start = max(vmf->address - nr * PAGE_SIZE,
>>> vma->vm_start);
>>> +    unsigned long end = min(start + folio_nr_pages(folio) *
>>> PAGE_SIZE, vma->vm_end);
>>> +    pte_t *start_ptep = vmf->pte - (vmf->address - start) / PAGE_SIZE;
>>> +    bool pte_write_upgrade = vma_wants_manual_pte_write_upgrade(vma);
>>> +    unsigned long addr;
>>> +
>>> +    /* Restore all PTEs' mapping of the large folio */
>>> +    for (addr = start; addr != end; start_ptep++, addr += PAGE_SIZE) {
>>> +        pte_t pte, old_pte;
>>> +        pte_t ptent = ptep_get(start_ptep);
>>> +        bool writable = false;
>>> +
>>> +        if (!pte_present(ptent) || !pte_protnone(ptent))
>>> +            continue;
>>> +
>>> +        if (vm_normal_folio(vma, addr, ptent) != folio)
>>> +            continue;
>>> +
>>
>> Should you be using folio_pte_batch() in the caller to collect all
>> applicable PTEs and then only have function that batch-changes a given
>> nr of PTEs?
>>
>> (just like we are now batching other stuff)
> 
> Seems folio_pte_batch() is not suitable for numa balancing, since we did
> not care about other PTE bits, only care about the protnone bits. And

You should be able to ignore most bits we care about, which case are you 
concerned about folio_pte_batch() would miss. Hand crafting own 
functions to cover some corner cases nobody cares about is likely a bad 
idea.

> after more thinking, I think I can drop the vm_normal_folio()
> validation, since all PTEs are ensured to be within the range of the
> folio size.

Are you sure about that?

-- 
Cheers,

David / dhildenb

Powered by blists - more mailing lists