| Message ID | 20220214200017.3150590-5-willy@infradead.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Various fixes around invalidate_page() | expand |
On Mon, Feb 14, 2022 at 08:00:11PM +0000, Matthew Wilcox (Oracle) wrote: > folio_mapped() is expensive because it has to check each page's mapcount > field. A cheaper check is whether there are any extra references to > the page, other than the one we own and the ones held by the page cache. > The call to remove_mapping() will fail in any case if it cannot freeze > the refcount, but failing here avoids cycling the i_pages spinlock. I wonder if something like this should also be in a comment near the check in the code.
On 2022/2/15 4:00, Matthew Wilcox (Oracle) wrote: > folio_mapped() is expensive because it has to check each page's mapcount > field. A cheaper check is whether there are any extra references to > the page, other than the one we own and the ones held by the page cache. > The call to remove_mapping() will fail in any case if it cannot freeze > the refcount, but failing here avoids cycling the i_pages spinlock. > > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> > --- LGTM. Thanks. Reviewed-by: Miaohe Lin <linmiaohe@huawei.com> > mm/truncate.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/truncate.c b/mm/truncate.c > index b73c30c95cd0..d67fa8871b75 100644 > --- a/mm/truncate.c > +++ b/mm/truncate.c > @@ -287,7 +287,7 @@ int invalidate_inode_page(struct page *page) > return 0; > if (folio_test_dirty(folio) || folio_test_writeback(folio)) > return 0; > - if (page_mapped(page)) > + if (folio_ref_count(folio) > folio_nr_pages(folio) + 1) > return 0; > if (folio_has_private(folio) && !filemap_release_folio(folio, 0)) > return 0; >
On Mon, Feb 14, 2022 at 11:19:22PM -0800, Christoph Hellwig wrote: > On Mon, Feb 14, 2022 at 08:00:11PM +0000, Matthew Wilcox (Oracle) wrote: > > folio_mapped() is expensive because it has to check each page's mapcount > > field. A cheaper check is whether there are any extra references to > > the page, other than the one we own and the ones held by the page cache. > > The call to remove_mapping() will fail in any case if it cannot freeze > > the refcount, but failing here avoids cycling the i_pages spinlock. > > I wonder if something like this should also be in a comment near > the check in the code. /* The refcount will be elevated if any page in the folio is mapped */ is what I've added for now.
On Mon, Feb 14, 2022 at 08:00:11PM +0000, Matthew Wilcox (Oracle) wrote: > folio_mapped() is expensive because it has to check each page's mapcount > field. A cheaper check is whether there are any extra references to > the page, other than the one we own and the ones held by the page cache. > The call to remove_mapping() will fail in any case if it cannot freeze > the refcount, but failing here avoids cycling the i_pages spinlock. This is the patch that's causing ltp's readahead02 test to break. Haven't dug into why yet, but it happens without large folios, so I got something wrong. > diff --git a/mm/truncate.c b/mm/truncate.c > index b73c30c95cd0..d67fa8871b75 100644 > --- a/mm/truncate.c > +++ b/mm/truncate.c > @@ -287,7 +287,7 @@ int invalidate_inode_page(struct page *page) > return 0; > if (folio_test_dirty(folio) || folio_test_writeback(folio)) > return 0; > - if (page_mapped(page)) > + if (folio_ref_count(folio) > folio_nr_pages(folio) + 1) > return 0; > if (folio_has_private(folio) && !filemap_release_folio(folio, 0)) > return 0; > -- > 2.34.1 >
On Fri, Feb 25, 2022 at 01:31:09AM +0000, Matthew Wilcox wrote: > On Mon, Feb 14, 2022 at 08:00:11PM +0000, Matthew Wilcox (Oracle) wrote: > > folio_mapped() is expensive because it has to check each page's mapcount > > field. A cheaper check is whether there are any extra references to > > the page, other than the one we own and the ones held by the page cache. > > The call to remove_mapping() will fail in any case if it cannot freeze > > the refcount, but failing here avoids cycling the i_pages spinlock. > > This is the patch that's causing ltp's readahead02 test to break. > Haven't dug into why yet, but it happens without large folios, so > I got something wrong. This fixes it: +++ b/mm/truncate.c @@ -288,7 +288,8 @@ int invalidate_inode_page(struct page *page) if (folio_test_dirty(folio) || folio_test_writeback(folio)) return 0; /* The refcount will be elevated if any page in the folio is mapped */ - if (folio_ref_count(folio) > folio_nr_pages(folio) + 1) + if (folio_ref_count(folio) > + folio_nr_pages(folio) + 1 + folio_has_private(folio)) return 0; if (folio_has_private(folio) && !filemap_release_folio(folio, 0)) return 0; Too late for today's -next, but I'll push it out tomorrow.
diff --git a/mm/truncate.c b/mm/truncate.c index b73c30c95cd0..d67fa8871b75 100644 --- a/mm/truncate.c +++ b/mm/truncate.c @@ -287,7 +287,7 @@ int invalidate_inode_page(struct page *page) return 0; if (folio_test_dirty(folio) || folio_test_writeback(folio)) return 0; - if (page_mapped(page)) + if (folio_ref_count(folio) > folio_nr_pages(folio) + 1) return 0; if (folio_has_private(folio) && !filemap_release_folio(folio, 0)) return 0;
folio_mapped() is expensive because it has to check each page's mapcount field. A cheaper check is whether there are any extra references to the page, other than the one we own and the ones held by the page cache. The call to remove_mapping() will fail in any case if it cannot freeze the refcount, but failing here avoids cycling the i_pages spinlock. Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> --- mm/truncate.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)