Re: [PATCH v2 2/4] mm/util: introduce snapshot_page()

[Luiz Capitulino <luizcap@redhat.com>]

On 2025-07-11 07:56, David Hildenbrand wrote:
> On 07.07.25 20:50, Luiz Capitulino wrote:
>> This commit refactors __dump_page() into snapshot_page().
>>
>> snapshot_page() tries to take a faithful snapshot of a page and its
>> folio representation. The snapshot is returned in the struct
>> page_snapshot parameter along with additional flags that are best
>> retrieved at snapshot creation time to reduce race windows.
>>
>> This function is intended to be used by callers that need a stable
>> representation of a struct page and struct folio so that pointers
>> or page information doesn't change while working on a page.
>>
>> The idea and original implementation of snapshot_page() comes from
>> Matthew Wilcox with suggestions for improvements from David Hildenbrand.
>> All bugs and misconceptions are mine.
>>
>> Signed-off-by: Luiz Capitulino <luizcap@redhat.com>
>> ---
>>   include/linux/mm.h | 19 ++++++++++++
>>   mm/debug.c         | 42 +++----------------------
>>   mm/util.c          | 77 ++++++++++++++++++++++++++++++++++++++++++++++
>>   3 files changed, 100 insertions(+), 38 deletions(-)
>>
>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>> index 0ef2ba0c667a..090968c6eebb 100644
>> --- a/include/linux/mm.h
>> +++ b/include/linux/mm.h
>> @@ -4184,4 +4184,23 @@ static inline bool page_pool_page_is_pp(struct page *page)
>>   }
>>   #endif
>> +#define PAGE_SNAPSHOT_FAITHFUL     (1 << 0)
>> +#define PAGE_SNAPSHOT_PG_FREE      (1 << 1)
> 
> Can we call this "BUDDY" instead of FREE
> 
> There are other types of free pages -- in particular free hugetlb folios -- which won't be covered by this check. We really only care about "buddy" pages, which correspond to "free" pages.
> 
> That's what we check and alter expose :)

OK, I'll change.

>> +#define PAGE_SNAPSHOT_PG_IDLE      (1 << 2)
>  > +> +struct page_snapshot {
>> +    struct folio folio_snapshot;
>> +    struct page page_snapshot;
>> +    unsigned long pfn;
>> +    unsigned long idx;
>> +    unsigned long flags;
>> +};
>> +
>> +static inline bool snapshot_page_is_faithful(const struct page_snapshot *ps)
>> +{
>> +    return ps->flags & PAGE_SNAPSHOT_FAITHFUL;
>> +}
>> +
>> +void snapshot_page(struct page_snapshot *ps, const struct page *page);
>> +
>>   #endif /* _LINUX_MM_H */
>> diff --git a/mm/debug.c b/mm/debug.c
>> index 907382257062..7349330ea506 100644
>> --- a/mm/debug.c
>> +++ b/mm/debug.c
>> @@ -129,47 +129,13 @@ static void __dump_folio(struct folio *folio, struct page *page,
>>   static void __dump_page(const struct page *page)
>>   {
>> -    struct folio *foliop, folio;
>> -    struct page precise;
>> -    unsigned long head;
>> -    unsigned long pfn = page_to_pfn(page);
>> -    unsigned long idx, nr_pages = 1;
>> -    int loops = 5;
>> -
>> -again:
>> -    memcpy(&precise, page, sizeof(*page));
>> -    head = precise.compound_head;
>> -    if ((head & 1) == 0) {
>> -        foliop = (struct folio *)&precise;
>> -        idx = 0;
>> -        if (!folio_test_large(foliop))
>> -            goto dump;
>> -        foliop = (struct folio *)page;
>> -    } else {
>> -        foliop = (struct folio *)(head - 1);
>> -        idx = folio_page_idx(foliop, page);
>> -    }
>> +    struct page_snapshot ps;
>> -    if (idx < MAX_FOLIO_NR_PAGES) {
>> -        memcpy(&folio, foliop, 2 * sizeof(struct page));
>> -        nr_pages = folio_nr_pages(&folio);
>> -        if (nr_pages > 1)
>> -            memcpy(&folio.__page_2, &foliop->__page_2,
>> -                   sizeof(struct page));
>> -        foliop = &folio;
>> -    }
>> -
>> -    if (idx > nr_pages) {
>> -        if (loops-- > 0)
>> -            goto again;
>> +    snapshot_page(&ps, page);
>> +    if (!snapshot_page_is_faithful(&ps))
>>           pr_warn("page does not match folio\n");
>> -        precise.compound_head &= ~1UL;
>> -        foliop = (struct folio *)&precise;
>> -        idx = 0;
>> -    }
>> -dump:
>> -    __dump_folio(foliop, &precise, pfn, idx);
>> +    __dump_folio(&ps.folio_snapshot, &ps.page_snapshot, ps.pfn, ps.idx);
>>   }
>>   void dump_page(const struct page *page, const char *reason)
>> diff --git a/mm/util.c b/mm/util.c
>> index 0b270c43d7d1..c38d213be83f 100644
>> --- a/mm/util.c
>> +++ b/mm/util.c
>> @@ -25,6 +25,7 @@
>>   #include <linux/sizes.h>
>>   #include <linux/compat.h>
>>   #include <linux/fsnotify.h>
>> +#include <linux/page_idle.h>
>>   #include <linux/uaccess.h>
>> @@ -1171,3 +1172,79 @@ int compat_vma_mmap_prepare(struct file *file, struct vm_area_struct *vma)
>>       return 0;
>>   }
>>   EXPORT_SYMBOL(compat_vma_mmap_prepare);
>> +
>> +static void set_flags(struct page_snapshot *ps, const struct folio *folio,
>> +              const struct page *page)
>> +{
>> +    /*
>> +     * Caveats on high order pages: PG_buddy and PG_slab will only be set
>> +     * on the head page.
>> +     */
> 
> Talking about slab here is a bit misleading, IIRC they are proper compound pages today. For PG_buddy (which was renamed to PGTY_buddy), we
> are dealing with non-compound higher allocations.
> 
> So talking about "head" is also a bit misleading.
> 
> /*
>   * Only the first page of a high-order buddy page has PageBuddy() set.
>   * So we have to check manually whether this page is part of a high-
>   * order buddy page.
>   */

I'll use this comment.

>> +    if (PageBuddy(page))
>> +        ps->flags |= PAGE_SNAPSHOT_PG_FREE;
>> +    else if (page_count(page) == 0 && is_free_buddy_page(page))
>> +        ps->flags |= PAGE_SNAPSHOT_PG_FREE;
>> +
>> +    if (folio_test_idle(folio))
>> +        ps->flags |= PAGE_SNAPSHOT_PG_IDLE;
>> +}
>> +
>> +/*
> 
> I would suggest writing proper kernel doc (you mostly have that already)
> 
> Something like
> 
> /**
>   * snapshot_page() - Create a snapshot of a "struct page"
>   * @ps: struct page_snapshot to store the page snapshot
>   * @page: the page we want to snapshot
>   *
>   * Create a snapshot of a page and store its struct page and struct
>   * folio ...

I'll fix this too.

> 
>> + * Create a snapshot of a page and store its struct page and struct folio
>> + * representations in a struct page_snapshot.
>> + *
>> + * @ps: struct page_snapshot to store the page snapshot
>> + * @page: the page we want to snapshot
>> + *
>> + * Note that creating a faithful snapshot of a page may fail if the page
>> + * compound keeps changing (eg. due to folio split). In this case we set
>> + * ps->faithful to false and the snapshot will assume that @page refers
>> + * to a single page.
>> + */
>