On 15.08.23 15:25, Daniel Gomez wrote:
Hi Yin,
On Tue, Aug 08, 2023 at 10:09:17AM +0800, Yin Fengwei wrote:
Commit 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a
folio") replaced the page_mapcount() with folio_mapcount() to check
whether the folio is shared by other mapping.
It's not correct for large folios. folio_mapcount() returns the total
mapcount of large folio which is not suitable to detect whether the folio
is shared.
Use folio_estimated_sharers() which returns a estimated number of shares.
That means it's not 100% correct. It should be OK for madvise case here.
I'm trying to understand why it should be ok for madvise this change, so
I hope it's okay to ask you few questions.
folio_mapcount() calculates the total maps for all the subpages of a
folio. However, the folio_estimated_sharers does it only for the first
subpage making it not true for large folios. Then, wouldn't this change
drop support for large folios?
It's all a mess right now.
1) page_mapcount(page): how often it this page mapped
For a THP: entire mapcount of the THP (PMD-mapping) + mapcount of *this
very subpage* (PTE-mapping) only
2) folio_mapcount(): how often is this folio mapped
For a THP: entire mapcount of the THP (PMD-mapping) + mapcount of *all*
subpages (PTE-mapping) of the folio
3) folio_estimated_sharers(): how often is the first page mapped
For a THP: entire mapcount of the THP (PMD-mapping) + mapcount of *the
first subpage* (PTE-mapping) only
For the time being, folio_estimated_sharers() is better then
folio_mapcount(), because for a PTE-mapped THP folio_mapcount() > 1.
I'm looking into a replacement for folio_estimated_sharers() that is
more precise ("folio_mapped_shared()"), but it's all a bit tricky. :)
--
Cheers,
David / dhildenb
