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

[Shivank Garg <shivankg@amd.com>]

On 7/14/2025 6:46 PM, Luiz Capitulino wrote:
> On 2025-07-08 12:59, Luiz Capitulino wrote:
>> On 2025-07-08 01:49, Shivank Garg wrote:
>>>
>>>
>>> On 7/8/2025 12:20 AM, 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)
>>>> +#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.
>>>> +     */
>>>> +    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;
>>>> +}
>>>> +
>>>> +/*
>>>> + * 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.
>>>> + */
>>>> +void snapshot_page(struct page_snapshot *ps, const struct page *page)
>>>> +{
>>>> +    unsigned long head, nr_pages = 1;
>>>> +    struct folio *foliop, folio;
>>>> +    int loops = 5;
>>>> +
>>>> +    ps->pfn = page_to_pfn(page);
>>>> +    ps->flags = PAGE_SNAPSHOT_FAITHFUL;
>>>> +
>>>> +again:
>>>> +    memcpy(&ps->page_snapshot, page, sizeof(*page));
>>>> +    head = ps->page_snapshot.compound_head;
>>>> +    if ((head & 1) == 0) {
>>>> +        foliop = (struct folio *)&ps->page_snapshot;
>>>> +        ps->idx = 0;
>>>> +        if (!folio_test_large(foliop)) {
>>>> +            set_flags(ps, page_folio(page), page);
>>>> +            goto out;
>>>> +        }
>>>> +        foliop = (struct folio *)page;
>>>> +    } else {
>>>> +        foliop = (struct folio *)(page->compound_head - 1);
>>>                       ^^^^
>>> should we use ps->page_snapshot here?
>>> IIUC, page may change due to race.
>>
>> You're right the race exists, but we can't make foliop point to the
>> snapshot because we do pointer arithmetic and other operations that need
>> the real page memory address to work properly.
> 
> While working on v3 I realized it should be fine (and correct) to use
> the head pointer from ps->page_snapshot like the original __dump_page()
> does (assuming this is what you meant for me to change). So, I changed
> this for v3.
> 
Yes, that's what I meant.
Thanks for updating it.

>>
>>>> +        ps->idx = folio_page_idx(foliop, page);
>>>> +    }
>>>> +
>>>> +    if (ps->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));
>>>> +        set_flags(ps, foliop, page);
>>>> +        foliop = &folio;
>>>> +    }
>>>> +
>>>> +    if (ps->idx > nr_pages) {
>>>> +        if (loops-- > 0)
>>>> +            goto again;
>>>> +        ps->page_snapshot.compound_head &= ~1UL;
>>>
>>> Should we use clear_compound_head() for clarity?
>>
>> Yes, we can do that.
>>
>> Andrew, since you applied this series already, can you squash the change
>> below into this patch? If not, I can send v3.
>>
>> diff --git a/mm/util.c b/mm/util.c
>> index c38d213be83f..ce4200529772 100644
>> --- a/mm/util.c
>> +++ b/mm/util.c
>> @@ -1239,7 +1239,7 @@ void snapshot_page(struct page_snapshot *ps, const struct page *page)
>>       if (ps->idx > nr_pages) {
>>           if (loops-- > 0)
>>               goto again;
>> -        ps->page_snapshot.compound_head &= ~1UL;
>> +        clear_compound_head(&ps->page_snapshot);
>>           foliop = (struct folio *)&ps->page_snapshot;
>>           ps->flags = 0;
>>           ps->idx = 0;
>