From: "Kirill A. Shutemov" <kirill@shutemov.name>
To: Matthew Wilcox <willy@infradead.org>
Cc: linux-mm@kvack.org, linux-fsdevel@vger.kernel.org,
Song Liu <songliubraving@fb.com>,
William Kucharski <william.kucharski@oracle.com>,
Johannes Weiner <jweiner@fb.com>
Subject: Re: [PATCH 3/3] mm: Allow find_get_page to be used for large pages
Date: Fri, 6 Sep 2019 15:59:28 +0300 [thread overview]
Message-ID: <20190906125928.urwopgpd66qibbil@box> (raw)
In-Reply-To: <20190905182348.5319-4-willy@infradead.org>
On Thu, Sep 05, 2019 at 11:23:48AM -0700, Matthew Wilcox wrote:
> From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
>
> Add FGP_PMD to indicate that we're trying to find-or-create a page that
> is at least PMD_ORDER in size. The internal 'conflict' entry usage
> is modelled after that in DAX, but the implementations are different
> due to DAX using multi-order entries and the page cache using multiple
> order-0 entries.
>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
> include/linux/pagemap.h | 9 +++++
> mm/filemap.c | 82 +++++++++++++++++++++++++++++++++++++----
> 2 files changed, 84 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index d2147215d415..72101811524c 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -248,6 +248,15 @@ pgoff_t page_cache_prev_miss(struct address_space *mapping,
> #define FGP_NOFS 0x00000010
> #define FGP_NOWAIT 0x00000020
> #define FGP_FOR_MMAP 0x00000040
> +/*
> + * If you add more flags, increment FGP_ORDER_SHIFT (no further than 25).
Maybe some BUILD_BUG_ON()s to ensure FGP_ORDER_SHIFT is sane?
> + * Do not insert flags above the FGP order bits.
> + */
> +#define FGP_ORDER_SHIFT 7
> +#define FGP_PMD ((PMD_SHIFT - PAGE_SHIFT) << FGP_ORDER_SHIFT)
> +#define FGP_PUD ((PUD_SHIFT - PAGE_SHIFT) << FGP_ORDER_SHIFT)
> +
> +#define fgp_order(fgp) ((fgp) >> FGP_ORDER_SHIFT)
>
> struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
> int fgp_flags, gfp_t cache_gfp_mask);
> diff --git a/mm/filemap.c b/mm/filemap.c
> index ae3c0a70a8e9..904dfabbea52 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -1572,7 +1572,71 @@ struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
>
> return page;
> }
> -EXPORT_SYMBOL(find_get_entry);
> +
> +static bool pagecache_is_conflict(struct page *page)
> +{
> + return page == XA_RETRY_ENTRY;
> +}
> +
> +/**
> + * __find_get_page - Find and get a page cache entry.
> + * @mapping: The address_space to search.
> + * @offset: The page cache index.
> + * @order: The minimum order of the entry to return.
> + *
> + * Looks up the page cache entries at @mapping between @offset and
> + * @offset + 2^@order. If there is a page cache page, it is returned with
Off by one? :P
> + * an increased refcount unless it is smaller than @order.
> + *
> + * If the slot holds a shadow entry of a previously evicted page, or a
> + * swap entry from shmem/tmpfs, it is returned.
> + *
> + * Return: the found page, a value indicating a conflicting page or %NULL if
> + * there are no pages in this range.
> + */
> +static struct page *__find_get_page(struct address_space *mapping,
> + unsigned long offset, unsigned int order)
> +{
> + XA_STATE(xas, &mapping->i_pages, offset);
> + struct page *page;
> +
> + rcu_read_lock();
> +repeat:
> + xas_reset(&xas);
> + page = xas_find(&xas, offset | ((1UL << order) - 1));
Hm. '|' is confusing. What is expectation about offset?
Is round_down(offset, 1UL << order) expected to be equal offset?
If yes, please use '+' instead of '|'.
> + if (xas_retry(&xas, page))
> + goto repeat;
> + /*
> + * A shadow entry of a recently evicted page, or a swap entry from
> + * shmem/tmpfs. Skip it; keep looking for pages.
> + */
> + if (xa_is_value(page))
> + goto repeat;
> + if (!page)
> + goto out;
> + if (compound_order(page) < order) {
> + page = XA_RETRY_ENTRY;
> + goto out;
> + }
compound_order() is not stable if you don't have pin on the page.
Check it after page_cache_get_speculative().
> +
> + if (!page_cache_get_speculative(page))
> + goto repeat;
> +
> + /*
> + * Has the page moved or been split?
> + * This is part of the lockless pagecache protocol. See
> + * include/linux/pagemap.h for details.
> + */
> + if (unlikely(page != xas_reload(&xas))) {
> + put_page(page);
> + goto repeat;
> + }
> + page = find_subpage(page, offset);
> +out:
> + rcu_read_unlock();
> +
> + return page;
> +}
>
> /**
> * find_lock_entry - locate, pin and lock a page cache entry
> @@ -1614,12 +1678,12 @@ EXPORT_SYMBOL(find_lock_entry);
> * pagecache_get_page - find and get a page reference
> * @mapping: the address_space to search
> * @offset: the page index
> - * @fgp_flags: PCG flags
> + * @fgp_flags: FGP flags
> * @gfp_mask: gfp mask to use for the page cache data page allocation
> *
> * Looks up the page cache slot at @mapping & @offset.
> *
> - * PCG flags modify how the page is returned.
> + * FGP flags modify how the page is returned.
> *
> * @fgp_flags can be:
> *
> @@ -1632,6 +1696,10 @@ EXPORT_SYMBOL(find_lock_entry);
> * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
> * its own locking dance if the page is already in cache, or unlock the page
> * before returning if we had to add the page to pagecache.
> + * - FGP_PMD: We're only interested in pages at PMD granularity. If there
> + * is no page here (and FGP_CREATE is set), we'll create one large enough.
> + * If there is a smaller page in the cache that overlaps the PMD page, we
> + * return %NULL and do not attempt to create a page.
Is it really the best inteface?
Maybe allow user to ask bitmask of allowed orders? For THP order-0 is fine
if order-9 has failed.
> *
> * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
> * if the GFP flags specified for FGP_CREAT are atomic.
> @@ -1646,9 +1714,9 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
> struct page *page;
>
> repeat:
> - page = find_get_entry(mapping, offset);
> - if (xa_is_value(page))
> - page = NULL;
> + page = __find_get_page(mapping, offset, fgp_order(fgp_flags));
> + if (pagecache_is_conflict(page))
> + return NULL;
> if (!page)
> goto no_page;
>
> @@ -1682,7 +1750,7 @@ struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
> if (fgp_flags & FGP_NOFS)
> gfp_mask &= ~__GFP_FS;
>
> - page = __page_cache_alloc(gfp_mask);
> + page = __page_cache_alloc_order(gfp_mask, fgp_order(fgp_flags));
> if (!page)
> return NULL;
>
> --
> 2.23.0.rc1
>
--
Kirill A. Shutemov
next prev parent reply other threads:[~2019-09-06 12:59 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-05 18:23 [PATCH 0/3] Large pages in the page cache Matthew Wilcox
2019-09-05 18:23 ` [PATCH 1/3] mm: Add __page_cache_alloc_order Matthew Wilcox
2019-09-05 18:58 ` Song Liu
2019-09-05 19:02 ` Matthew Wilcox
2019-09-05 19:06 ` Song Liu
2019-09-05 18:23 ` [PATCH 2/3] mm: Allow large pages to be added to the page cache Matthew Wilcox
2019-09-05 18:28 ` Matthew Wilcox
2019-09-05 20:56 ` kbuild test robot
2019-09-06 12:09 ` Kirill A. Shutemov
2019-09-06 13:31 ` Matthew Wilcox
2019-09-05 18:23 ` [PATCH 3/3] mm: Allow find_get_page to be used for large pages Matthew Wilcox
2019-09-05 21:41 ` kbuild test robot
2019-09-05 22:04 ` kbuild test robot
2019-09-05 22:12 ` Matthew Wilcox
2019-09-09 0:42 ` [kbuild-all] " Rong Chen
2019-09-09 1:12 ` Matthew Wilcox
2019-09-06 12:59 ` Kirill A. Shutemov [this message]
2019-09-06 13:41 ` Matthew Wilcox
2019-09-06 13:52 ` Kirill A. Shutemov
2019-09-06 15:22 ` Matthew Wilcox
2019-09-06 15:59 ` [PATCH 4/3] Prepare transhuge pages properly Matthew Wilcox
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20190906125928.urwopgpd66qibbil@box \
--to=kirill@shutemov.name \
--cc=jweiner@fb.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=songliubraving@fb.com \
--cc=william.kucharski@oracle.com \
--cc=willy@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).