Received-SPF: pass (google.com: domain of linux-kernel+bounces-142584-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Message-ID: <2f6cc7a4-ea64-40aa-842a-8d85309a5cbd@arm.com>
Date: Fri, 12 Apr 2024 12:31:07 +0100
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v2 4/5] mm: swap: entirely map large folios found in
 swapcache
Content-Language: en-GB
To: Barry Song <21cnbao@gmail.com>
Cc: akpm@linux-foundation.org, linux-mm@kvack.org,
 baolin.wang@linux.alibaba.com, chrisl@kernel.org, david@redhat.com,
 hanchuanhua@oppo.com, hannes@cmpxchg.org, hughd@google.com,
 kasong@tencent.com, surenb@google.com, v-songbaohua@oppo.com,
 willy@infradead.org, xiang@kernel.org, ying.huang@intel.com,
 yosryahmed@google.com, yuzhao@google.com, ziy@nvidia.com,
 linux-kernel@vger.kernel.org
References: <20240409082631.187483-1-21cnbao@gmail.com>
 <20240409082631.187483-5-21cnbao@gmail.com>
 <1008d688-757a-4c2d-86bd-793f5e787d30@arm.com>
 <CAGsJ_4wqXFcS=WnfTSuS1osPh-NGtUF_izA2U3VVEv3wEhsudA@mail.gmail.com>
From: Ryan Roberts <ryan.roberts@arm.com>
In-Reply-To: <CAGsJ_4wqXFcS=WnfTSuS1osPh-NGtUF_izA2U3VVEv3wEhsudA@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

On 12/04/2024 00:30, Barry Song wrote:
> On Fri, Apr 12, 2024 at 3:33 AM Ryan Roberts <ryan.roberts@arm.com> wrote:
>>
>> On 09/04/2024 09:26, Barry Song wrote:
>>> From: Chuanhua Han <hanchuanhua@oppo.com>
>>>
>>> When a large folio is found in the swapcache, the current implementation
>>> requires calling do_swap_page() nr_pages times, resulting in nr_pages
>>> page faults. This patch opts to map the entire large folio at once to
>>> minimize page faults. Additionally, redundant checks and early exits
>>> for ARM64 MTE restoring are removed.
>>>
>>> Signed-off-by: Chuanhua Han <hanchuanhua@oppo.com>
>>> Co-developed-by: Barry Song <v-songbaohua@oppo.com>
>>> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
>>> ---
>>>  mm/memory.c | 64 +++++++++++++++++++++++++++++++++++++++++++----------
>>>  1 file changed, 52 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index c4a52e8d740a..9818dc1893c8 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -3947,6 +3947,10 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>       pte_t pte;
>>>       vm_fault_t ret = 0;
>>>       void *shadow = NULL;
>>> +     int nr_pages = 1;
>>> +     unsigned long start_address = vmf->address;
>>> +     pte_t *start_pte = vmf->pte;
>>
>> possible bug?: there are code paths that assign to vmf-pte below in this
>> function, so couldn't start_pte be stale in some cases? I'd just do the
>> assignment (all 4 of these variables in fact) in an else clause below, after any
>> messing about with them is complete.
>>
>> nit: rename start_pte -> start_ptep ?
> 
> Agreed.
> 
>>
>>> +     bool any_swap_shared = false;
>>
>> Suggest you defer initialization of this to your "We hit large folios in
>> swapcache" block below, and init it to:
>>
>>         any_swap_shared = !pte_swp_exclusive(vmf->pte);
>>
>> Then the any_shared semantic in swap_pte_batch() can be the same as for
>> folio_pte_batch().
>>
>>>
>>>       if (!pte_unmap_same(vmf))
>>>               goto out;
>>> @@ -4137,6 +4141,35 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>        */
>>>       vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
>>>                       &vmf->ptl);
>>
>> bug: vmf->pte may be NULL and you are not checking it until check_pte:. Byt you
>> are using it in this block. It also seems odd to do all the work in the below
>> block under the PTL but before checking if the pte has changed. Suggest moving
>> both checks here.
> 
> agreed.
> 
>>
>>> +
>>> +     /* We hit large folios in swapcache */
>>> +     if (start_pte && folio_test_large(folio) && folio_test_swapcache(folio)) {
>>
>> What's the start_pte check protecting?
> 
> This is exactly protecting the case vmf->pte==NULL but for some reason it was
> assigned in the beginning of the function incorrectly. The intention of the code
> was actually doing start_pte = vmf->pte after "vmf->pte = pte_offset_map_lock".
> 
>>
>>> +             int nr = folio_nr_pages(folio);
>>> +             int idx = folio_page_idx(folio, page);
>>> +             unsigned long folio_start = vmf->address - idx * PAGE_SIZE;
>>> +             unsigned long folio_end = folio_start + nr * PAGE_SIZE;
>>> +             pte_t *folio_ptep;
>>> +             pte_t folio_pte;
>>> +
>>> +             if (unlikely(folio_start < max(vmf->address & PMD_MASK, vma->vm_start)))
>>> +                     goto check_pte;
>>> +             if (unlikely(folio_end > pmd_addr_end(vmf->address, vma->vm_end)))
>>> +                     goto check_pte;
>>> +
>>> +             folio_ptep = vmf->pte - idx;
>>> +             folio_pte = ptep_get(folio_ptep);
>>> +             if (!is_swap_pte(folio_pte) || non_swap_entry(pte_to_swp_entry(folio_pte)) ||
>>> +                 swap_pte_batch(folio_ptep, nr, folio_pte, &any_swap_shared) != nr)
>>> +                     goto check_pte;
>>> +
>>> +             start_address = folio_start;
>>> +             start_pte = folio_ptep;
>>> +             nr_pages = nr;
>>> +             entry = folio->swap;
>>> +             page = &folio->page;
>>> +     }
>>> +
>>> +check_pte:
>>>       if (unlikely(!vmf->pte || !pte_same(ptep_get(vmf->pte), vmf->orig_pte)))
>>>               goto out_nomap;
>>>
>>> @@ -4190,6 +4223,10 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>                        */
>>>                       exclusive = false;
>>>               }
>>> +
>>> +             /* Reuse the whole large folio iff all entries are exclusive */
>>> +             if (nr_pages > 1 && any_swap_shared)
>>> +                     exclusive = false;
>>
>> If you init any_shared with the firt pte as I suggested then you could just set
>> exclusive = !any_shared at the top of this if block without needing this
>> separate fixup.
> 
> Since your swap_pte_batch() function checks that all PTEs have the same
> exclusive bits, I'll be removing any_shared first in version 3 per David's
> suggestions. We could potentially develop "any_shared" as an incremental
> patchset later on .

Ahh yes, good point. I'll admit that your conversation about this went over my
head at the time since I hadn't yet looked at this.

> 
>>>       }
>>>
>>>       /*
>>> @@ -4204,12 +4241,14 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>        * We're already holding a reference on the page but haven't mapped it
>>>        * yet.
>>>        */
>>> -     swap_free(entry);
>>> +     swap_free_nr(entry, nr_pages);
>>>       if (should_try_to_free_swap(folio, vma, vmf->flags))
>>>               folio_free_swap(folio);
>>>
>>> -     inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
>>> -     dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
>>> +     folio_ref_add(folio, nr_pages - 1);
>>> +     add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr_pages);
>>> +     add_mm_counter(vma->vm_mm, MM_SWAPENTS, -nr_pages);
>>> +
>>>       pte = mk_pte(page, vma->vm_page_prot);
>>>
>>>       /*
>>> @@ -4219,33 +4258,34 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>        * exclusivity.
>>>        */
>>>       if (!folio_test_ksm(folio) &&
>>> -         (exclusive || folio_ref_count(folio) == 1)) {
>>> +         (exclusive || (folio_ref_count(folio) == nr_pages &&
>>> +                        folio_nr_pages(folio) == nr_pages))) {
>>>               if (vmf->flags & FAULT_FLAG_WRITE) {
>>>                       pte = maybe_mkwrite(pte_mkdirty(pte), vma);
>>>                       vmf->flags &= ~FAULT_FLAG_WRITE;
>>>               }
>>>               rmap_flags |= RMAP_EXCLUSIVE;
>>>       }
>>> -     flush_icache_page(vma, page);
>>> +     flush_icache_pages(vma, page, nr_pages);
>>>       if (pte_swp_soft_dirty(vmf->orig_pte))
>>>               pte = pte_mksoft_dirty(pte);
>>>       if (pte_swp_uffd_wp(vmf->orig_pte))
>>>               pte = pte_mkuffd_wp(pte);
>>
>> I'm not sure about all this... you are smearing these SW bits from the faulting
>> PTE across all the ptes you are mapping. Although I guess actually that's ok
>> because swap_pte_batch() only returns a batch with all these bits the same?
> 
> Initially, I didn't recognize the issue at all because the tested
> architecture arm64
> didn't include these bits. However, after reviewing your latest swpout series,
> which verifies the consistent bits for soft_dirty and uffd_wp, I now
> feel  its safety
> even for platforms with these bits.

Yep, agreed.

> 
>>
>>> -     vmf->orig_pte = pte;
>>
>> Instead of doing a readback below, perhaps:
>>
>>         vmf->orig_pte = pte_advance_pfn(pte, nr_pages);
> 
> Nice !
> 
>>
>>>
>>>       /* ksm created a completely new copy */
>>>       if (unlikely(folio != swapcache && swapcache)) {
>>> -             folio_add_new_anon_rmap(folio, vma, vmf->address);
>>> +             folio_add_new_anon_rmap(folio, vma, start_address);
>>>               folio_add_lru_vma(folio, vma);
>>>       } else {
>>> -             folio_add_anon_rmap_pte(folio, page, vma, vmf->address,
>>> -                                     rmap_flags);
>>> +             folio_add_anon_rmap_ptes(folio, page, nr_pages, vma, start_address,
>>> +                                      rmap_flags);
>>>       }
>>>
>>>       VM_BUG_ON(!folio_test_anon(folio) ||
>>>                       (pte_write(pte) && !PageAnonExclusive(page)));
>>> -     set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
>>> -     arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte);
>>> +     set_ptes(vma->vm_mm, start_address, start_pte, pte, nr_pages);
>>> +     vmf->orig_pte = ptep_get(vmf->pte);
>>> +     arch_do_swap_page(vma->vm_mm, vma, start_address, pte, pte);
>>>
>>>       folio_unlock(folio);
>>>       if (folio != swapcache && swapcache) {
>>> @@ -4269,7 +4309,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
>>>       }
>>>
>>>       /* No need to invalidate - it was non-present before */
>>> -     update_mmu_cache_range(vmf, vma, vmf->address, vmf->pte, 1);
>>> +     update_mmu_cache_range(vmf, vma, start_address, start_pte, nr_pages);
>>>  unlock:
>>>       if (vmf->pte)
>>>               pte_unmap_unlock(vmf->pte, vmf->ptl);
>>
> 
> Thanks
> Barry